institute f'or defense analysesnonparametric statistical techniques are used to circumvent...

ID A INSTITUTE F'OR DEFENSE ANALYSES

An Empirical Examination of CounterdrugInterdiction Program Effectiveness

Dr. Barry D. Crane, Project LeaderDr. A. Rex Rivolo, Principal Analyst

Dr. Gary C. Comfort

January 1997

Approved for public release;distribution unlimited.

IDA Paper P-3219

Log: H 96-003599

19970203 025

This work was conducted under contract DASWO1 94 C 0054, TaskA-173, for the Deputy Assistant Secretary of Defense, Drug EnforcementPolicy and Support. The publication of this IDA document does not indi-cate endorsement by the Department of Defense, nor should the con-tents be construed as reflecting the official position of that Agency.

© 1997 Institute for Defense Analyses, 1801 N. Beauregard Street,Alexandria, Virginia 22311-1772 * (703) 845-2000.

This material may be reproduced by or for the U.S. Government pursuantto the copyright license under the clause at DFARS 252.227-7013(10/88).

INSTITUTE FOR DEFENSE ANALYSES

IDA Paper P-3219

An Empirical Examination of CounterdrugInterdiction Program Effectiveness

Dr. Barry D. Crane, Project LeaderDr. A. Rex Rivolo, Principal Analyst

Dr. Gary C. Comfort

PREFACE

This paper was produced by the Institute for Defense Analyses (IDA) for the

Deputy Assistant Secretary of Defense, Drug Enforcement Policy and Support,

Department of Defense in response to the task Counterdrug Research on the Effectiveness

of Detection, Monitoring, and Interdiction in the Western Hemishpere.

The IDA Technical Review Committee was chaired by Mr. Thomas P. Christie and

consisted of the following IDA reviewers: Dr. Arthur Fries, Dr. David R. Graham,

Dr. Dale E. Lichtblau, Dr. Bernard H. Paiewonsky, Dr. Paul H. Richanbach, and

Dr. Richard H. White. This final version also benefits from comments provided by several

qualified reviewers external to IDA.

The authors are indebted to Professor Tyler Cowen, Economics Department,

George Mason University, for his review and assistance in identifying and describing

several hidden costs caused by interdiction.

111°°

AN EMPIRICAL EXAMINATION OF COUNTERDRUG

INTERDICTION PROGRAM EFFECTIVENESS

Table of Contents

S u m m ary .............................................................................................................. 1

I. B ackgrou nd ............................................................................................... I-I

II. The Pricing Structure of the Cocaine Market ..................... II-1

A . B ackground ......................................................................................... II-1B . The STRID E D ata Base ....................................................................... 11-2C. Empirical Evidence for a Multiplicative Cocaine Market ...................... II-10D. Theoretical Foundation for a Multiplicative Market ............... 11-13E. Extracting Intrinsic Time Dependence in Price and Purity ..................... 11-18

III. The Relationship Between Cocaine Price and Usage .............................. III-1

IV. The Effects of Interdiction Campaigns on the Cocaine Market ............. IV-1

A. Timing of Interdiction Events .............................................................. IV-1B. The Long-Term Benefits of Interdiction ............................................... IV-7C. The Cost-Effectiveness of Interdiction ................................................. IV-13

1. Cost-Effectiveness of Source-Zone Interdiction ............................... IV-132. Cost-Effectiveness of Total Interdiction ........................................... IV-14

V. Reconciling Empirically Derived Estimates of Cost-Effectivenesswith Previously Published Estimates ...................................................... V-1

A. Inadequate and Incomplete M odel ........................................................ V-2B. Assumption of an Additive Pricing Structure ........................................ V-5C. Unreliable Estimates of Prevalence Derived from the Demand Model... V-6D. Extrapolation of the Marginal Cost of Additional New Treatments ....... V-10E. Inclusion of Inappropriate Costs In Determining the Costs of Source-

Country Control and Interdiction ........ ...... .............. V-12F. Summary of Cost-Effectiveness Estimates ............................................ V-13

V I. F in d in gs ..................................................................................................... V I-1

R eferences .............................................................................................................. R -1

Appendix A - Acronyms

v

List of Figures

1. Price history of the U.S. cocaine market with superimposed time markersshowing the timing of all major source-zone interdiction events .................. 2

II-la. Gross purchase volume distribution for all STRIDE purchases between1980 and mid-1996 in the range of 0 to 20 grams ........................................ I-4

II-lb. Gross purchase volume distribution for all STRIDE purchases between1980 and mid-1996 in the range of 0 to 150 grams ..................... 11-4

II-lc. Gross purchase volume distribution for all STRIDE purchases between1980 and mid-1996 in the range of 100 to 600 grams .................................. I-5

IH- Id. Gross purchase volume distribution for all STRIDE purchases between1980 and mid-1996 in the range of 400 to 2,400 grams ............................... 1-5

11-2. Distribution of normalized purchase volume for all STRIDE purchasesbetween 1980 and mid-1996 in the range 0 to 20 grams .............................. 11-6

II-3a. Normalized unit price distribution for all STRIDE purchases between 1980and mid-1996 within the normalized purchase volume range of 0 gm to 1.0g m .............................................................................................................. 11 -7

II-3b. Normalized unit price distribution for all STRIDE purchases between 1980and mid-1996 within the normalized purchase volume range of 1.0 gm to10 g m ......................................................................................................... 1 -7

11-3c. Normalized unit price distribution for all STRIDE purchases between 1980and mid-1996 within the normalized purchase volume range of 10 gm to10 0 g m ....................................................................................................... I -8

II-3d. Normalized unit price distribution for all STRIDE purchases between 1980and mid-1996 within the normalized purchase volume range of 100 gm to10 k g .......................................................................................................... 11 -8

11-4. Probability of paying more than x $/gm when purchasing cocaine on thestreet as determined from the STRIDE data base for three ranges inpurchase volum e ......................................................................................... 11-10

11-5. Median normalized unit prices for cocaine for three market distributionlevels ...................................................................................................... .. 11 -11

11-6. Supply and demand curves illustrating the nonlinear behavior resultingfrom supply interdiction .............................................................................. I11-14

1H-7. Median normalized unit price versus normalized purchase volume for allSTRIDE data between 1980 and mid-1996 .................................................. 11-17

11-8. Price history of the cocaine market as defined by the street price index ........ 11-2011-9. Purity history of the cocaine market as defined by the median purity of the

STR ID E sam ple .......................................................................................... 11-20III-1. Correlation between the STRIDE-derived cocaine street price index and the

number of DA WTN emergency room cocaine mentions .................................. 111-4111-2. Correlation between the STRIDE-derived cocaine street price index and the

D UF cocaine positive rate ........................................................................... 111-5

vi

111-3. Correlation between the STRIDE-derived cocaine street price index and theSmithKline Beecham Clinical Laboratories combined workforce cocainepositive test rate .......................................................................................... 111-6

111-4. Correlation between the STRIDE-derived cocaine street price index and thepercentage of TEDS treatment center cocaine admissions ............................ 111-7

IV-1. Price history of the U.S. cocaine market with superimposed time markersshowing the timing of all major source-zone interdiction events since 1980. IV-4

IV-2. Estimates of coca base transport on the air bridge from Peru to Colombia... IV-5IV-3. History of coca base price and estimated cost in Peru .................................. IV-7IV-4. Time history of STRIDE median gross purchase unit-price from 1983 to

m id-19 9 6 .................................................................................................... IV -9IV-5. Street price index of cocaine since 1985 ...................................................... IV-1 1IV-6. Detailed view of the street price index since 1989 ........................................ IV-1 1IV-7. History of median normalized unit-price and median purity for all STRIDE

purchases in the range (0 and 10 grams) of normalized purchase volumesince 1992 ................................................................................................... IV -12

V-1. Result of two fits to the same data for cocaine prevalence ........................... V-8V-2. History of Federally-funded drug treatments and funding expenditures ........ V-1IV-3. Comparison of program cost-effectiveness estimates ................................... V-13

List of Tables

III-1. Summary of Price Elasticity Estimates for Various Indicators of CocaineU sag e ......................................................................................................... 111-8

vii

SUMMARY

This paper presents an analysis of much of the available data relating to the cocaine

industry and from that analysis arrives at four primary findings: 1) significant upward

excursions in cocaine street prices have occurred in the U.S. since 19831; 2) U.S.

interdiction efforts in conjunction with support from the source nations have been the

likely cause of these excursions; 3) the price excursions have resulted in measurable

reductions in cocaine usage within the U.S.; 4) a strategy of using interdiction external to

the U.S. to disrupt cocaine market dynamics and thus raise prices can be a relatively cost-

effective approach to reducing cocaine use.

The paper arrives at its conclusions by examining extensive data on the price and

purity of cocaine sold in the U.S., as reported by the Drug Enforcement Administration's

System to Retrieve Information from Drug Evidence (STRIDE); operational data from

various U.S. and source nation counter-drug agencies; and a variety of indirect or

secondary drug-use indicators derived from various sources. These sources of indirect

usage indicators include: the Drug Abuse Warning Network (DA WN) hospital emergency

room data; the Drug Usage Forecast (DUF) data derived from drug testing of arrested

felons; the Treatment Episode Data Set (TEDS) derived from Government treatment

centers; and the drug testing program conducted by SmithKline Beecham Clinical

Laboratories (SBCL) for civilian and Government organizations.

Nonparametric statistical techniques are used to circumvent serious but poorly

understood idiosyncrasies of the STRIDE data and to construct a cocaine street price

index2 reflecting the intrinsic time variation in cocaine prices within the U.S. market as a

whole. This price index reveals that the time history of cocaine prices since 1983 is

characterized by a smoothly decreasing trend that abruptly changed in 1989, and on which

are superimposed a number of distinct, short duration, price excursions or "bumps." The

"Street price" as used in this paper is not synonomous with "retail price" as commonly usedelsewhere.

2 The street price index is defined as the median normalized unit price for the aggregate of the entire

STRIDE sample. Justification for this definition is presented in Chapter II.

sudden change in the price decay rate and each of the short-term excursions are shown to

follow the initiation of major interdiction 3 activities, primarily in the source nations, and

are thus to be causally connected. Figure 1 shows the cocaine street price index history

since 1983 and the timing of the eight major source-zone interdiction events conducted by

the U.S. during that time interval. The open circles are individual price index

determinations; the solid line through the data is the result of a smoothing technique

applied to mitigate statistical noise.

Timing of Interdiction Events350 , . ......... .. .

:1. OP. CHEMCON

1 :2. OP. BLAST FURN4ACE I

:3. OP. BLAST FURNACE II- 30 0 ... .... : .. ...... !.......: ........ i ....... !. ..... . . .. ...... i b h :w N • b : • C i " " i......

.- 0

:5. OP. S UPPORT JUSTC E I I0• o i6. OP. sOPPoR , JO's,,CE T,,VIo' | o ... 7. ,OP SILVA VERDE ((CNP)'

00

250C .20 . ...o" o!..... ...o....... ...... ....... ........ ..........................-.. ...........'E'ENT'O~ i....

,: . o.' N/ - .? - : : : : : : 8. :HO W POIC IM LMNE:200 • 0

7< 02 0 .. .. 0 ° • o . ..0 . . . .:. . . ."....... " ....... : ....... :....... :........ : ....... : ....... ". . .

q , I .. ........ . .. .X )

~00C:- 150 .. . ' '... ...15 . .... .:....... :...... • " o....... ....... I ....... : ... : ....... I ....... : ........ :. ...... - 1 . ...... : . . .

IT:)

16 : 500 0 '.,, o ....,, ,o : 9 .10 . ....... :. . .. .. ........: ........ .. .. " o0 ..i ..... o ...... ... ° : 6 . . o.. ........ .... .... .. . ..... . .......0 0. .. %: %- : •: o

V 5o0 ....... :- ....... :....... 0 0 ..... :.... .. ?.. ... .... ................... ..9 %

83 84 85 86 87 88 89 90 91 92 93 94 95 96 97Time (Year)

Figure 1. Price history of the U.S. cocaine market with superimposed time markersshowing the timing of all major source-zone interdiction events.

The paper then proceeds to show that when cocaine prices have increased

noticeably, the four independent drug-use indicators have decreased proportionally,

indicating some measure of reduced cocaine consumption. The relation between drug

price and drug use is shown to be consistent with the canonical value for the price

elasticity of demand (-0.5) commonly used by researchers in the field. Finally, using

estimates for total U.S. expenditures in international interdiction and the canonical value

for the price-demand elasticity, a rough estimate of cost-effectiveness indicates that the

The term "interdiction" used in this paper has a broad scope - those activities exterior to the borders

of the U.S. to prevent the production and transport of raw materials and coca products.

2

cost of decreasing cocaine use by one percent through the use of source zone interdiction

efforts is on the order of a few tens of millions of dollars and not on the order of a billion

dollars as reported in previous research. The differences are primarily attributed to a

failure in the earlier research to account for the major costs imposed on the traffickers by

interdiction operations and overestimation of the costs of conducting interdiction

operations.

The paper is organized as follows:

Chapter I provides a general background. It identifies the manner in which IDA

was tasked to examine interdiction activities, the scope of the tasking, and the manner in

which the IDA study team approached that tasking.

Chapter II examines the STRIDE data base and presents the methodology used to

derive the intrinsic time dependence of prices through the use of an IDA-defined streetprice index. The nature of the STRIDE data base is documented, with a focus on

distributional idiosyncrasies that make the use of the customary multi-parameterregression techniques problematic. Justification is provided for aggregating STRIDE data

at all purchase volumes into a single street price index that characterizes the domestic

marketplace.

Chapter III examines several indirect or secondary indicators of drug usage in theU.S. and shows that these are all inversely correlated with the street price index.

Chapter IV examines the time dependence of the street price index and argues thatinterdiction activities, primarily source-zone actions, have been the likely cause of the

excursions and the abrupt change in slope seen in 1989. This chapter further considers

whether the excursions in price, attributed to source-zone interdiction actions, result in

long-term structural changes in the market, and presents a rough estimate for the cost-

effectiveness of interdiction.

Finally, Chapter V addresses the contrast between the cost-effectiveness found in

this study and that of a previously published and widely distributed estimate. The

differences in the estimates are attributed primarily to five areas of inappropriate

assumption or methodology in the previous study.

3

CHAPTER I

BACKGROUND

I. BACKGROUND

Counterdrug interdictioni efforts are routinely grouped into two broad categories,referred to as source-zone interdiction and transit-zone interdiction. The source zone

category refers to activities in the primary coca growing nations, notably Peru and Bolivia,

and the transportation lanes from these countries to Colombia, where cocaine sulfate,commonly referred to as "base," is refined into the final product, cocaine hydrochloride,which is marketed in the United States. The transit zone refers to the transportation

routes for this final product from South America to the United States. The transit zoneincludes the Caribbean area, Central America, Mexico, and the adjacent Pacific Ocean

area. Along with the Department of Defense (DoD), numerous other agencies routinelyplay a vital role in the interdiction process, including the U.S. Coast Guard, the Drug

Enforcement Administration (DEA), the U.S. Customs Service, intelligence agencies, andthe military and police agencies of the source and transit zone nations.

As part of a major shift in U.S. counterdrug objectives, President's DecisionDirective 14, issued in 1993, redirected effort and resources from transit-zone to sourcecountry interdiction and host nation support. Paradoxically, some policy research results,

current at that time, argued that source country control and host nation support were theleast cost effective of all drug control strategies. As part of IDA's responsibilities to

evaluate U.S. detection, monitoring, and interdiction for our sponsors (the Department ofDefense and the U.S. Interdiction Coordinator), new research approaches were initiated to

determine the effectiveness of interdiction and methods to improve operations.

At the U.S. Quarterly Joint Staff/U.S. Interdiction Coordinator Interagency

Conference in December 1995, IDA was tasked by the U.S. interagency community to

examine the relationships between cocaine market price and interdiction activities.

Consistent with its operational orientation and experience, the IDA study team adopted anapproach of collecting and examining the extensive operational data bases describing

actual drug trafficking and usage experience. Such data include known and suspected

drug trafficker routings and flight tracks from the source-zone countries through the

The term "interdiction" used in this paper has a broad scope - those activities exterior to the bordersof the U.S. to prevent the production and transport of raw materials and coca products.

I-1

transit zone, drug price and purity data maintained by the DEA, and compilations of

statistical data collected by the U:S. Department of Health and Human Services related to

drug use (prevalence and consumption).

One of the seminal cocaine data bases is that maintained by the Drug Enforcement

Administration as the System to Retrieve Information from Drug Evidence (STRIDE)program. The STRIDE data base (DEA, 1994-1996) includes the quantity, price, purity,

and purchase location of actual or negotiated street buys by DEA agents. After applyingnonparametric data processing techniques2 to the STRIDE data, the IDA team noted that

cocaine street prices since 1980 are characterized by a smoothly decreasing trend onwhich are superimposed a number of distinct, short duration, upward excursions or

"bumps." Furthermore, these perturbations appeared to correspond in time with theinitiation of several specific source-zone interdiction activities. This led to IDA's

hypothesis that properly focused activities in the source-zone countries aimed at supply

disruption can produce significant excursions in the price and purity of cocaine in the U.S.

IDA presented this hypothesis to the interdiction community at the May 1995Quarterly Interagency Conference. At this time, we strongly urged focused attention onthe "air bridge" linking Peru and Colombia as an obvious "weak link" in the industry. Atthe December 1995 quarterly conference, IDA estimated that the shoot-down policyinitiated by Peru in March 1995 and the resulting disruption of the industry should haveled to a 30 to 50 percent increase in the cocaine street prices. When the STRIDE data for

the relevant time period became available (in late December 1995), the estimates were

confirmed by analysis of data and were corroborated by independent sources (Krauss,

1995). This result increased IDA's confidence that the correlations observed between

source zone events and street price excursions were consistent with a causal relationship.

In January 1996, the U.S. Interdiction Coordinator was briefed regarding thiscorrelation and conveyed a sense of opportunity for capitalizing on this market disruption

mechanism. The Interdiction Coordinator indicated his belief that the results weresignificant and called a special meeting of The Interdiction Committee (TIC) to be briefed

on our findings. TIC members recommended that IDA collect the next few months ofSTRIDE data for further confirmation before sponsoring the presentation of these new

results to the Office of National Drug Control Policy (ONDCP).

2 A detailed explanation of the nonparametric techniques used and the rationale for their use, ratherthan parametric methods, is presented in Chapter II.

1-2

At approximately the same time, the DoD office sponsoring this study arranged astaff level discussion with key staff from ONDCP. That discussion raised a question

concerning the marketplace linkage between price changes and changes in drug usage.

Consequently, IDA examined several data bases that are logically related to drug usage:

the SmithKline Beecham Clinical Labs (SBCL) data on random drug testing in theworkplace; the Department of Health and Human Services Drug Warning Network

(DAWN) data base on emergency room treatments; the Treatment Episode Data System(TEDS) data base on public drug treatment programs; and the Department of Justice Drug

Usage Forecasting (DUF) data base from quasi-random testing of arrestees in major cities.

Nonparametric analytical techniques were again employed to extract trend datawith the maximum possible time resolution. Applying formal linear regression techniques

to binned data produced significant correlations between all of the indirect usageindicators and the STRIDE price changes, indicating that cocaine use is responsive to price

changes.

This research showed significant effects on street prices directly affecting cocaineusage resulting from several source zone initiatives that had been conducted with relativelymodest expenditures. This finding - that source-zone interdiction can be relatively cost-

effective - was in direct conflict with the findings of a previously published study entitled

"Controlling Cocaine" (Rydell and Everingham, 1994), which found source-zone

activities to be the least cost-effective of the four cocaine control strategies considered.That study concluded that the most cost-effective of those strategies, treatment of heavy

users, was some 23 times more cost-effective in reducing cocaine usage than a source-

zone interdiction strategy. Consequently, the sponsor of this research requested that our

analytical effort be expanded to identify the cause of the apparent large differences

between our findings regarding the effectiveness of source zone interdiction efforts and

those presented in Controlling Cocaine. Accordingly, IDA undertook a review of

Controlling Cocaine and its companion report Modeling the Demand for Cocaine(Everingham and Rydell, 1994). Our review of these reports identified several assumptions

and methodologies that appeared to produce modeled results in disagreement with the

actual data.

The above-outlined analyses were documented in a draft IDA document, An

Empirical Examination of Counterdrug Interdiction Program Effectiveness (Crane, et al.,1996). That document-was intended to convey initial findings to those involved in the

operational management of the drug interdiction efforts in a timely manner as part of alimited distribution of the quarterly conference proceedings, while encouraging review and

1-3

comments from other researchers. Subsequently, a number of substantive comments wereobtained both from additional reviewers internal to IDA as well as from externalreviewers. Reviewers questioned the methodology, assumptions, definitions, and some of

the findings themselves.

This paper provides an expanded discussion of the key findings of the draft IDAdocument and attempts to resolve expressed concerns of methodology, definition, and

assumptions. In addition, care is taken to identify explicitly those findings considered to

be of primary importance. This paper continues to include judgments and interpretations

by the authors with which others may disagree, but by detailing our supporting data,

methodology and logic, it is hoped that any remaining disagreements with other

researchers in this area may be clarified.

1-4

CHAPTER II

THE PRICING STRUCTURE OF THE COCAINE MARKET

II. THE PRICING STRUCTURE OF THE COCAINE MARKET

This chapter develops a suitable statistic from the STRIDE data base with which to

quantify intrinsic changes in cocaine market prices. This statistic is then used to

demonstrate that the commonly assumed additive pricing structure for the cocaine market

is inconsistent with the empirical evidence. Under the additive model, costs imposed by

interdiction upon the production and/or distribution network would have insignificant

effect upon the retail price of cocaine. Rather, the empirical evidence indicates that such

imposed costs can have a significant effect on the price of cocaine at all subsequent levels

of the distribution network. The intrinsic time dependence of cocaine price and purity is

presented along with some conceptual hypotheses for why the cocaine market behaves as

observed.

A. BACKGROUND

Since the early days of the modem cocaine epidemic (1970s), the prevailing viewhas been that interdiction efforts have little effect on drug use prevalence or consumption

(see: Reuter, Crawford, and Cave, 1988; Cave and Reuter, 1988; Caulkins and Padman,

1993; Riley, 1993). The principal reason for this conclusion appears to be the general

belief that the cocaine market is characterized by an "additive" pricing structure. Under

the additive model, production and operational price increases at the producer or

wholesale level are added on to the retail price of the product. Since production and other

import costs are only a small percentage of the final retail price for cocaine, under an

additive model, such changes in cost have a negligibly small effect on the final price. The

alternative view, which is supported by the available data, suggests a "multiplicative"

pricing structure in which cost increases in the earlier levels of the cocaine distribution

hierarchy result in significant price effects at all levels of the market.

Reuter and Kleiman (1986) were among the first to model the cocaine market as

an additive model. Using the limited data available at that time, they analyzed many

plausible price-affecting mechanisms and concluded that "the enforcement-oriented

strategy will not work." Since that work was published, the basic conceptual framework

has not changed despite the increasing availability of additional market data that has led to

increased doubts concerning the validity of the additive model assumption. For example,

II-1

Caulkins (1993) finds evidence that "lends circumstantial support to the multiplicative

model" and concludes that the available data "are not consistent with the additive model."

Caulkins further concludes that if the multiplicative model is valid for the cocaine market,

"then, contrary to conventional wisdom, interventions and policy changes that affect

wholesale prices would be expected to significantly affect retail prices." Additional

discussion on this point can be found in Boyum (1992).

This study finds that the cocaine market is multiplicative in nature and not additive.Both the quantity of the data analyzed and the analytical techniques used to interpret those

data account for the differences between our results and earlier studies. In this work,

nonparametric regression techniques are used to extract time dependence, whereas most

other researchers have relied on parametric multiple regression techniques or on

"standardizing" of individual observations, both of which require explicit assumptions of

dependencies, and both of which are error-prone because of the pathological nature of thedistributions involved. Although parametric models can correctly account for alldependencies within the data, discovering the correct model, or formulating it from firstprinciples, is problematic. Dealing with the pathologies of the data in a parametric

approach is exceedingly difficult.

B. THE STRIDE DATA BASE

The STRIDE database contains the purchase time, ti, purchase quantity, Qi,purchase price, Pi, purchase purity, ir/, and purchase location for approximately 40,000

undercover purchases of cocaine since 1980. (Throughout this paper, the term volume

will be used synonymously with "quantity" and all prices have been inflation adjusted to

1992 dollars.) The STRIDE purchases range in volumes from 100 milligrams to tens ofkilograms.

Some discussion of how accurately the STRIDE sample reflects the true market

sale volume distribution is warranted. As a first approximation, one might assume that

STRIDE purchases are a random sampling of the "contracts" available in the market.

After all, a DEA agent seeking to make a purchase must go through the same process as

any user/dealer looking to make a buy. However, this assumption might be a distortedview. Agents might be targeting larger volume dealers preferentially, which may distort

the shape of the true purchase volume distribution. Alternatively, agents might be subject

to a "first purchase" surcharge not routinely experienced by the established buyer, therebybiasing the price. Whatever agents are doing, however, has little relevance to the

conclusions drawn in this paper. Since this paper focuses upon changes evident in the

11-2

STRIDE data, the only requirement is that the STRIDE sampling be approximately

stationary in time, (i.e., rate of purchase, and the shape of the purchase volume

distribution are approximately constant), a characteristic that statistical tests show to be

true for STRIDE data after 1985. Although some data prior to 1980 exist, the quality of

these data is questionable according to the DEA, and we have chosen not to use them.

From 1980 to 1985 a systematic bias toward lower volume purchases is clearly evident.

This systematic bias makes a direct comparison of absolute price values between widely

separated times questionable if pre-1985 data are involved, but on short time scales (a few

years) or post-1985, a direct comparison is valid.

Figures II-la to II-ld show the distribution of STRIDE gross (uncorrected for

purity) purchase volumes for all purchases from 1980 to mid-1996 in four parts: 0 to 20

grams, 0 to 150 grams, 100 to 600 grams, and 400 to 2,400 grams, respectively. (Figure

II-I is presented in four parts only for ease of publication, visualization, and viewing

different scale sizes, and should be considered as a single continuous figure. The overlap

of ranges for the four parts is intended to aid the reader in mentally connecting the parts.)

The counts in the histograms are per unit purchase volume, i.e., the purchase volume bin

sizes vary among the four plots; the number of counts in each bin has been divided by the

width of the bin to make the ordinate independent of bin size and allow the four plots to

be compared directly to each other.1

Within each observed range, the purchase volume distribution is seen to be highly

skewed toward low values with charaateristic peaks at the preferred market transaction

values (e.g., 1/2 oz, 1 oz, 2 oz, ... 1/2 kg, 1 kg). This distribution suggests that themarket naturally divides into three groupings - one group at the kilogram level (1/8, 1/4,

1/2, 1, 2, 3 kg, etc.), one group at the ounce level (1/8, 1/4, 1/2, 1, 2, 3, 4 oz) (note: 1 oz= 28 grams), and one group at the gram level - essentially distributed as a hyperbolically

or exponentially decaying continuum. A similar division of the market into gram, ounce,

and kilogram groupings was adopted by Caulkins (1994) in his examination of STRIDE

data.

In analyzing the STRIDE data, it will be useful to define a unit-price, U, = P /Qiand two "normalized" quantities: the normalized purchase volume, (QN)i - riQ,, and the

Because some features of the data are narrower than the bin size, the independence of the ordinatewith bin size is only approximate. As a result, the height of a feature in one plot may be slightlydifferent than the same feature in a second plot.

11-3

STRIDE Gross Purchase Volume Distribution6000

5000 "

E_ 4 0 0 0 ., .. . . . . . . . . . .".. . .. . . . . . . . . . . . . . .. . . . . . - . . . . . . . . . ... . .

3 4000 .................. ........................... ............................

Q)

0~

:3:

12 0 0 0 ......... .. .. .. .. .. ............ ......... .................. .. .. .. .. . .. .. . ... ...0

1000 ....

0

0 5 10 15 20

Gross Purchase Volume, 0 (gm)

Figure lI-la. Gross purchase volume distribution for all STRIDE purchases between 1980and mid-1996 in the range of 0 to 20 grams.


4000 ................................................................. ...........

E

0

13000 .. ... . .... . ............. ... ........ ............................

0-

10 0 0 . .. .. . .. . . .. . .. . .. . . .. . .. . .. . . .. . .. . .. . .

0 25 50 75 100 125 150

Gross Purchase Volume, 0 (gm)

Figure lI-lb. Gross purchase volume distribution for all STRIDE purchases between 1980and mid-1996 in the range of 0 to 150 grams.

"H-4


S15 0 .. .. . .. . . .. . ; . .. . . .. . . .. .. . . . ............... ..... . .. .. .. .. ... .. ....... .. .. .. .. .. .. .. .. .. .. .S150 ....E

6

100 .100 ................ ................ ........ ............. ......................

00 5 0 . . . . .. . . ........ .. . .. ... . .. .. . . .. .

0

100 200 300 400 500 600Gross Purchase volume, 0 (gin)

Figure 11-1c. Gross purchase volume distribution for all STRIDE purchases between 1980and mid-1998 in the range of 100 to 800 grams.


20 .......... ' ..................... ...... ............ ....................... ... ........

E::

o> 15 ......... ......... .......... n .......... ,.......... .......... .. .. .. . .......... . .. . . ..........

5 ........ ... .................... . .... . .................0 -4 60 0 800 10014000 1800 2000 2200 2400

Gross Purchose Volume, Q (gin)

Figure Il-Id. Gross purchase volume distribution for all STRIDE purchases between 1980and mid-1996 in the range of 400 to 2,400 grams.

5-5

normalized unit-price, (UN)i Pi/(QN)i. The distinction between retail, middle, and

wholesale levels is quickly lost when one considers only the pure cocaine content of a

purchase transaction. For example, Figure 11-2 shows the same range of purchase volume

distribution as Figure II-la, but using the "normalized" volume, that is, the purchase

volume corrected to 100 percent pure cocaine content. Notice that all the features present

in the gross purchase volume distribution have been obliterated in the normalized

distribution, leaving only a smooth exponential-like continuum. This is the true market

sale volume distribution as sampled by DEA undercover agents.

STRIDE Normalized Purchase Volume Distribution10000

S7500 ......................................... ...........................................E73

o 500

0

0 5 10 15 20Normalized Purchase Volume, 0 N (gm)

Figure 11-2. Distribution of normalized purchase volume for all STRIDE purchasesbetween 1980 and mid-1996 in the range 0 to 20 grams.

From the normalized volume distribution shown in Figure 11-2, a normalized unit-

price distribution can be derived by dividing each actual purchase price by the normalized

purchase volume. Figures 11-3a to 11-3d show normalized purchase price distributions for

four purchase volume ranges. Again, all STRIDE purchases since 1980 are included. In

all histograms, the last bin in the plot contains all counts greater than the plot-limit value.

In the upper right-hand comer of each plot are the distribution inclusive volume range and

the 'distribution statistics. All the distributions have very long "tails" that extend further as

purchase volumes get smaller (e.g., notice the difference between the mean and the median

for each distribution).

11-6

STRIDE Normalized Purchase Unit-Price Distribution300

0 gmn 0 ,, < 1.0 gm

mean = 5208.52 5 0 . . .............. ..................... ....... .................................... & ; .. - L ~ ........ ...medan = 243.8N 5997

2O00

20 ..50 ........ .... .................................... ...................... ....................

U

100 .. .......... ....... .................................................

50

0 500 1000 1500 2000 2500Normalized Unit-Price, UN ($/gM)

Figure 11-3a. Normalized unit price distribution for all STRIDE purchases between 1980and mid-1996 within the normalized purchase volume range of 0 gm to 1.0 gin. The

rightmost bin contains all values greater than $2,500 per gram.


1.0 gm < Q, ( 10 gmmean = 171.8

400 ..................................... median..=....N = 12983

3 0 0 . ............................................................................ .

UCa)

Q) 200---------........ ........................................................

100......... ............................. ................................ ......................

00 250 500 750 1000

Normalized Unit-Price, UN ($/gm)

Figure 11-3b. Normalized unit price distribution for all STRIDE purchases between 1980and mid-1996 within the normalized purchase volume range of 1.0 gm to 10 gm. The

rightmost bin contains all values greater than $1,000 per gram.

H1-7


10 gm < . ( 100 gm

mean = 75.9

median = 63.0

N = 18985

500 ........................ ..... ............................ .. .... .............

ETC

Q)

250 . . .................. . .......................................................

00 100 200 300 400

Normalized Unit-Price, UN ($/gm)

Figure 11-3c. Normalized unit price distribution for all STRIDE purchases between 1980and mid-1996 within the normalized purchase volume range of 10 gm to 100 gm. The

rightmost bin contains all values greater than $400 per gram.


1100: gm < 0

N ( 10 kg

: mean = 36.1

100 ................................................. ............. ...

N N 2210

C)

C:a)

G) 50.......

25 . .......................... ................... .................. ..................

250 50 100 150 200 250

Normalized Unit-Price, UN (/gnm)

Figure 11-3d. Normalized unit price distribution for all STRIDE purchases between 1980and mid-1996 within the normalized purchase volume range of 100 gm to 10 kg. The

rightmost bin contains all values greater than $250 per gram.

11-8

The distribution of Figure II-3d is bimodal with a distinct component below $10

per gram. These purchases may reflect contracts made at the import price by DEA agents

dealing directly with importers or may reflect some other idiosyncrasy of the data.

Regardless of the origin of these low-priced purchases, they constitute a small fraction of

the total and do not affect the median values, used exclusively in this study, and therefore

do not greatly affect any statistical outcome.

In all cases, the long tail of the unit-price distribution is due partly to a strong

dependence between unit-price and purchase volume (the bigger the buy, the lower the

unit-price), but a long tail is also an intrinsic property of the unit price distribution, even

for narrowly restricted volume ranges. This tail arises because cocaine is almost always

adulterated, thus making the unit price of the actual cocaine content for any individual

purchase essentially a random variable. Since the degree of adulteration is strongly

dependent on the purchase volume, with larger volume purchases being observed to be

much purer and having smaller variances than smaller purchase volumes, a larger purchase

volume means a smaller variance.

The long tails of the normalized purchase unit price distributions obey the so-called

Pareto-Levy law in economics (Mandelbrot, 1983, and references therein). The tails

illustrate essentially asymptotic behavior in which the cumulative distribution function, for

large values of the independent variable, converges to an inverse power-law. If the

logarithmic slope of the power-law is shallower than -1.0, then the first and second

moments of the distribution diverge for an infinite sampling and are random noise for any

finite sampling. If the logarithmic slope is between -1.0 and -2.0, the first moment

converges, but the higher ones do not. It is the existence of these Pareto-Levy tails in the

price distributions that makes analysis of these data problematic.

Figure 11-4 shows explicitly the Pareto-Levy tails of the normalized unit-price

distributions for three of the four ranges of normalized purchase volume shown in Figure

11-3. The graph shows the probability of paying more than x dollars per gram as a

function of x. Notice that, for the 0 to 1 gram sales, the slope of the tail of the probability

curve is very close to -1.0; therefore, not only are the higher moments of this distribution

divergent, but so is the mean. The main consequence of this is that the mean price of

cocaine for purchases in the volume range of 0 to 1 gram is dominated by random noise.

This result calls into question all analytical studies that monitor prices by computing

quarterly (or longer) averages predominantly in the range of 0 to 1 gram. In addition,

11-9

Probability of Paying More Than x $/gm

•. ... .... . ...... .. .......... .......

o :

a-0

..

:0 - 1.0 gm-

1: 0! 10 9M

10 -100 grn

101 102 103 104 105

Normalized Unit Price, x ($,/gm)

Figure 11-4. Probability of paying more than x $/gm when purchasing cocaine on thestreet as determined from the STRIDE data base for three ranges in purchase volume.

because the slopes of the distribution tails are shallower than -2 for a large portion (in

particular the "retail" regions) of the STRIDE data, it also raises serious questions about

the validity of the ubiquitous multiple linear regression techniques that have been used to

extract price time dependencies or price-demand elasticities, since the methodology

underlying such techniques is predicated on the existence of finite variances.

C. EMPIRICAL EVIDENCE FOR A MULTIPLICATIVE COCAINE MARKET

In order to quantify market dynamics, it is necessary to extract the time

dependence of price and purity from the time series, that is, construct some statistic to

reflect intrinsic price and purity as a function of time. Before constructing such a statistic,

it is instructive to look at the time dependence of prices for narrow ranges in purchase

volume.

Because of the difficulties raised by the long Pareto-Levy tails of the unit-price

distributions, it is necessary to select a statistic insensitive to the tails with which to form a

time dependent function. Statistical principles indicate the use of the median value, rather

than the mean, since the robustness of this statistic is well documented. Figure 11-5 shows

II-10

a plot of the median normalized unit-price for three purchase volume intervals of the

STRIDE data between January 1983 and June 1996.2 The data have been "cut" into these

three purchase volume intervals to reflect the three notional market levels: retail (0-10

gram), middle (10-30 gram), and wholesale (> 30 gram), although, as previously pointed

out, such a distinction is, in our opinion, somewhat arbitrary.

Time Dependence of Normalized Unit-Price for 0-10 gm Sales400 * 0

3 , 5 ' 8 5o9 % 0 ; 9 3 9 9

T o N N gm1644S0E 300 0- . ........ "...

01.. . . . . I

83 54 85 55 87 85 89 go 91 92 93 94 95 96 " 97

Time Dependence of Normalized Unit--Price for Sals--30 gmSoe

00

2 0 0 ... .. .. . . . .. . .. .. ..

Z _ 0

0S00 0

0, 25 0 -

o i i i i I I i I I i i I

83 84 85 86 87 85 59 90 91 92 93 94 95 96 97

Time Dependence of Normalized Unit--Price for the maet stiu

150 g m .e 10 to 30 grams botom. 30grmsane............r.................

• ,, 00 . . . :: "0 . o o ..o)Z 75 ...... o -...... - _ - -- -

2 t figu r ..... i t l i....... . .... .....o .....tm depe d .... e ..... s r.....ti year is s as. 98

rahe thn18 bcueincuiono9018 dat gral copese th scl oftepoi hu

83 84 85 86 87 as 8g go 91 92 93 94 g5 96 97

addin gp-e. of Normalized Ugiainrt n.s c r for Salevels.15 1I1-1 1

Top: 0 to 0grm;mdl:1to3grm;bto:0grmanlagrThvlusoN125 eac .......... .... --------- ---------- -N. -77-28- -uc a e -ot i e in ea h g ou i g

2 nti fgr ndi h floigfiue f iedpednistesatigya i eeteos18rahe ..... ..... ...us inclusio ....... ...........al c m ressth caeoftepltwi h u

adding~ sinfcn ifrain

Z _ -~ ~ ~ ~~~~~I-- 1 1- -. .................

In this figure, and in all time sequences to follow, each open circle represents the

median value of L individual transactions that have been sorted to be consecutive in time

and within the appropriate volume range. The solid line running through the data points is

the result of a nonparametric regression technique, closely related to the well-known

median filtering approach (see: Mosteller and Tukey, 1977). The line is generated by

convolving the median data points with a triangular function, whose width is typically a

few months, to smooth out noise fluctuations. The procedure is essentially a weighting

scheme that replaces the jth data point by a weighted sum of K nearest neighbors

(bidirectionally) with the weights assigned in direct proportion to the time separation

between the data point and the neighbors. Since this approach is used repeatedly

throughout this paper, the procedure is explicitly described below.

Starting from the individual STRIDE purchase samples, the normalized unit price

of each transaction is computed.

(UN) --

The data are then sorted in time (ti) to form a time series, sectored into groups of L

samples each, and for each sector, a group median value is computed to define the median

normalized unit price, Z(tj), for the data index interval (j-J)L < i < jLjL

Z(tj) = X50 [ (UWii--(i-I)L

where X50 is a notation for the median operator. The time tj to which the median value Z

is assigned is the median of the times 1i over the interval. The "regression" function curve

is then given by the convolution+K12F(,j I Z(tj÷ ) W(t )

k=-K12

where W(tk) is a symmetric triangular weighting function with zero mean and unity

integral. The width of the base of this function (smoothing time. interval) is set by the

value of K. Note that none of the conclusions of this paper depend on the regression

curves derived by the above procedure. The regression curves are included solely to

enhance the exposition and help in visualization of the time dependence of the noisy data.

For the solid lines in Figure 11-5, the values of L and K have been set to L = 100

and K = 8. Notice that the three time-dependent curves in the figure are statistically

similar, that is, they all go up and down in direct proportion to each other. Slicing the data

into other purchase volume intervals always produces similar results, indicating that the

11-12

systematic local maxima and minima in the curves are not the result of sampling

fluctuations, but the result of intrinsic variations in time that affect all prices at various

stages of distribution simultaneously and proportionally to each other. This constitutes the

primary evidence supporting the conclusion of a multiplicative market.

The finding that the empirical evidence is consistent with a multiplicative model for

the cocaine pricing at all levels of the domestic market - and inconsistent with an additive

pricing model - argues that costs imposed at an early level of the cocaine production and

distribution network, e.g., as a result of interdiction activities, can produce much largerincreases in the retail price of cocaine than customarily has been recognized.

D. THEORETICAL FOUNDATION FOR A MULTIPLICATIVE MARKET

The argument that the effects of interdiction can yield significant increases in theretail price of cocaine receives support from several quarters. We first show that

significant movements in the street price of cocaine are consistent with standard economic

theory, and then discuss some other factors that may be special to the cocaine market and

that may amplify such price movements, even beyond what standard theory would

suggest.

Advocates of the additive pricing model typically argue that increased costs

imposed upon cocaine production or distribution by interdiction activities can produce

only (negligibly) small changes in the retail price of cocaine. The argument runs as

follows: the production costs for 1 gram of cocaine typically run a few dollars, whereas

the retail price is much higher, typically $100 to $200. A doubling of production cost, say

from $2/gm to $4/gm, therefore will affect the retail price only slightly, according to this

view, since the added $2/gram is only a small fraction of final market price.

This example poses a misleading dilemma for an economic analysis of interdiction.

Interdiction typically creates non-linear effects on costs. In other words, it is incorrect toclaim that marginal cost has risen a constant percentage across the entire range of the cost

curve. For some units, the marginal cost of production does not change at all. For other

units, the marginal cost of production becomes extremely high, perhaps infinite, at least in

the short run. Such non-linear effects are easy to see in the costs associated with

laboratory destruction, air interdiction, or spraying. Much of the market is not affected,

while other parts of the market are affected significantly. These effects create a sharp

upward break in the cost curve beyond a certain range.

11-13

The additive model assumes that interdiction acts simply to raise the costs of

cocaine coming into the U.S. But in addition to raising costs, interdiction also limits the

flow of cocaine supplies into the U.S. The nonlinear effects on supply is illustrated in

Figure 11-6, where interdiction shifts the supply curve from S to S'. Because the demand

for cocaine is relatively inelastic, such a reduction in the flow of supply may significantly

increase the short-run market-clearing price from P to P'. This is the same kind of price

spike that is often observed in other markets when the flow of supplies is disrupted, and

we believe this effect explains the significant price excursions observed in the data. In thelong-run, cocaine suppliers will adapt to market disruptions by adopting alternative

distribution channels, but they likely will not be able to bring costs down to the original

supply curve S, and there may continue to be supply-limiting effects even in the long-run.

Because the additive model overlooks these supply-limiting effects of interdiction, it

defines away a potent effect on cocaine pricing and usage.

Nonlinear Shift in Supply Curve

S1-)

0

D

C

Relative Supply (S) and Relative Demand (D)

Figure 11-6. Supply and demand curves illustrating the nonlinear behavior resulting fromsupply interdiction.

As noted in the previous section, we have observed a consistent multiplicative

pricing relationship in the domestic cocaine distribution system. This has persisted over

wide variations in prices, suggesting that the relationship is deeply ingrained in the cocaine

distribution culture. It may also reflect accepted distribution cost relationships. In an

11-14

illegal market such as this, the costs of the raw materials are small compared with the

costs associated with security, prosecution risk premiums, and other distribution costs at

the various tiers in the distribution system. At the small volume retail level, the costs of

security and the higher risks of arrest may be the dominant costs of business. The stability

of pricing relationships over time is consistent with a product having a low elasticity of

demand, and with a rigidly established marketing hierarchy. The economics of the

structure and pricing within the cocaine distribution system is worthy of further study.

Specifically, cocaine sale and distribution may be characterized by self-similar

hierarchical, or "fractal," structures. We offer the following remarks as an interesting lead

for future research, but emphasize that our basic conclusions do not depend upon the

acceptance or rejection of specific propositions about fractal market structure.

A fractal structure results when, at every level of the distribution chain, dealers buy

in quantity Q and resell in lots of quantity Q/M while applying a price markup of S. If M

and S are constant throughout the market, then the distribution chain is self-similar and a

fractal structure results. The parameter M is called the multiplicity or branching ratio and

the parameter S, the markup or price multiplier.

The analysis of fractal market structure is consistent with the standard framework

of tax incidence theory. If a product has many stages of production and distribution,

fractal results will emerge if a given percentage increase in cost has a more than

proportional increase on price at each production stage (on how prices respond to

increases in cost, see Atkinson and Stiglitz 1980, especially Lecture 7). Assume, for

instance, that a given percent increase in cost at the level of coca leaf production creates a

10 percent increase in the price of leaves. This implies a 10 percent increase in cost for

the next level of production, which in turn can increase the price of that stage's outputs

more than 10 percent. If the outputs of that stage rise in price by 20 percent, then the next

stage experiences a cost increase of 20 percent, and so on. The cost increase at each stage

increases the price of the inputs purchased by subsequent stages of production, and can

have a multiplicative effect on final market price, hence the name "multiplicative market

structure." The fractal concept refers to the ability of the price change to propagate itself

and expand through successive stages of production. Such a market structure requires

11-15

that the gross purchase price, P, as a function of transaction size, Q, satisfy the following

relation:

MP(QM) = SP(Q).

This implies a market in which volume discounts (unit-price -volume relation) are related

by a power-law

U(Q) OQr,

where U(Q) = P(Q)IQ. The exponent, y, in this relation is related to M and S by the

fundamental fractal relation:

= - log(S) /log(M).

Such an inverse power-law relation between unit purchase price and purchase volume is

indeed observed within the STRIDE data. This relation is valid for both the normalized

and gross quantities and is fundamental to this market. Figure HI-6 presents, on a log-log

scale, a plot of the median normalized purchase price from the STRIDE data base versus

the normalized purchased quantity (volume). Each point on Figure II-7 represents the

median of the normalized unit-prices of 100 transactions in the same normalized volume

bin. (For purchase volumes greater than 1,000 grams, each data point represents the

median of only 10 transactions.) The straight-line relationship between log price and log

quantity in the figure has a slope of -1/3 and spans about four orders of magnitude in

purchase volume. This power-law relationship is consistent with a self-similar hierarchical

distribution network model and is direct evidence for a fractal structure of the cocaine

market and the multiplicative nature of its pricing scheme.

Such a self-similar hierarchical market model was proposed by Caulkins and

Padman (1993). Although their model is not couched in the language of fractals, it does

contain all the necessary elements. A very similar fractal market structure may exist in the

heroin industry (although the STRIDE heroin data have not been yet examined). Moore

(1970), building on work by Preble and Casey (1967) who used information obtained from

dealers and users, outlines a near-hierarchical distribution chain for heroin with six levels.

With minor modifications, the system outlined by Moore can be transformed into a fractal

scheme with multiplicity M--10 and mark-up S=3.

11-16

Purchase Price-Volume Relation

0'

00:'0 °o

000

00 0

a-

0 0: 0

S

. .. .0 0

D 0

'o_00

10-2 10-1 100 101 102 103 104

Normalized Purchase Volume, QN (gin)

Figure 11-7. Median normalized unit price versus normalized purchase volume for allSTRIDE data between 1980 and mid-1996. The solid line through the data is Y = 200 X-M .

While a fractal structure may not be commonplace for normal, i.e., legal markets, it

is conceptually reasonable for an illegal market in which risk is the dominant force

mitigating profits. The risks to a dealer of being arrested or exposed to violence increase

as the multiplicity or branching ratio, M, increases, thereby providing an incentive to hold

down the number of customers, while profits, because of the inverse power-law between

unit-price and volume, increase rapidly as the multiplicity increases. Aggregating the data

into time intervals quickly shows that the exponent of the price-volume relation has been

constant since 1980, indicating that the interplay between risk and profit has not changed

significantly since that time.

The observed value of the exponent of the price-volume relation from the STRIDE

data (Figure 11-6) is approximately r= -1/3. The discussion in Section B suggested that

the street market naturally divides into three groupings at the kilogram, ounce, and gram

levels, respectively, thus revealing a preference for a multiplicity of approximately 30.

With an exponent r = -1/3, and M = 30, one calculates a price mark-up of about

S=3.

11-17

The interpretation and consequences of this fractal market structure are profound.

They imply a market that is multiplicative in price structure and that the product is sold

through a distribution network that is self-similar and hierarchical; that is, every dealer

takes approximately the same risk and makes approximately the same return on

investment. We are aware of no fractal markets other than this one, although fractal

structures are common in socio-economic systems (see: Mandelbrot, 1983, for a

discussion and references).

With the postulation of this fractal market structure, it is possible to construct a

reasonable outline of the market hierarchy in the United States. Consider the following

first-order estimate. Approximately 200 high-level dealers purchase lots of approximately

30 kilograms at an import price of approximately $7 per gram. These 200 dealers sell in

lots of 1 kilogram to 6,000 individuals at a unit price of about $20 per gram. These

dealers in turn sell in lots of 1 ounce (30 grams) to 180,000 individuals at $60 per gram,

who in turn sell in lots of 1 gram, at a unit price of $180 per gram, to 5.4 million end

users. The schematic below may help to visualize the hierarchy:

30 kg @ 7 $/gm => 200 := 1 kg @ 20 $/gm

1 kg @ 20 $/gm => 6,000 => 1 oz @ 60 $/gm

I oz @60 $/gm =* 180,000 =, 1 gm @ 180 $/gm

I gm @ 180 $/gm, => 5,400,000.

This schematic is not meant to be interpreted in detail. In reality, all of the

numbers are approximate, and, furthermore, they do not represent single values but the

integrals of distributions that may be near-normally distributed or highly skewed. In order

for the above scheme to account for a total consumption of 250 to 300 metric tons of

cocaine per year, the hierarchy sequence must be repeated an average of once per week.

E. EXTRACTING INTRINSIC TIME DEPENDENCE IN PRICE AND PURITY

For any range of purchase volume, a time-dependent function using the median of

the price distribution values can be formed. However, when the volume range is narrow,

there are few data points available and the function is dominated by statistical fluctuations.

To decrease the statistical noise, we use the observation that unit prices at all levels of the

market (all purchase volumes) are proportional to each other and that the variances are

very large. This allows' the aggregation of all purchases, irrespective of purchase volume,

to define a price index that reflects a price characteristic of the market as a whole without

11-18

distinction to wholesale, middle, or retail levels. Such an approach may appear to be

questionable since there are wide differences between high-level prices and low-levelprices. However, because all prices are proportional and have very large dispersions, the

composite median is itself proportional to all other prices, and thus, with greatly decreased

statistical noise, this composite median does an excellent job of resolving temporal

changes related to interdiction operations.

With the above justification, we therefore define a cocaine street price index as themedian normalized unit price for the aggregate of the entire STRIDE sample (all purchase

volumes), as:

SPI(t) = X 50 [ (UN)j]i=(j-1)L

typically with L = 100, but any reasonable binning size will yield similar results. Since

source zone interdiction operations generally affect the U.S. market as a whole, all

purchase locations are aggregated together. (Individual locations could be segregated atthe cost of greatly increased statistical noise.) Note that the "street price" as measured by

this index is not synonymous with the "retail price" often used in the literature, since theretail price normally refers to only the smallest volume transactions available on the street.

This street price index represents the unit price per pure gram of cocaine for the

median purchase volume of the entire STRIDE sample; we interpret this index as

characteristic of the cocaine market as a whole. In computing this index, care was takento ensure that no major changes in the purchase volume distribution have occurred over

the period of interest. As previously noted, because there is some systematic change in theSTRIDE purchase volume distribution before 1985, absolute values of this index cannot be

compared between two widely separated times if data prior to 1985 are included. Over

the short time-scale prior to 1985, or post-1985, however, the index is well-behaved andany systematic trends can be interpreted as real excursions. The great advantage of thisprocedure, of course, is that it uses all the data points to compute a single characteristic

price of the market as a whole, while minimizing statistical fluctuations and enhancing the

time resolution of the available data.

We turn now to the goal of the long discussion above - the intrinsic timedependence of cocaine price and purity. Figure 11-8 shows the street price index (inconstant 1992 dollars) as a function of time, and Figure 11-9 shows the corresponding

median purity. In both plots, L, the number of data points over which the medians are

computed, has been set to 100 and the smoothing function width at K = 8. The two figures

11-19

Cocaine Street Price History350 ... ....

S30 0 . .. .... .. .. .. .. ......... ..... .... .... . .. . . .. . .. .. , ....... ... . ., . ...... • . .. . .... .. ... .. .. .. ... . . . . . . .

E .0

2 5 0 2 ... .. 9 .. .. ....... ........ ....... ........ ........ ...... ..... . . . .. . . . . . . . .. . . . . . . .

00

CF) 200 000 150 -

00 ........ ... . .. .. .

50 .......... . . . . . . . .. .

20 .... ... " --- " : :0.- -- 0- ..............................--....--..... .................... ....... .....

0) .0 IFiur : Prc hsor of th cain make as deine by th stetpic indx

900

: 1500 . . .:. . . . .9 -........ ... . .: ....... : -- . ...... :. ... 6 6 .. .. . -:. ...o~~ .• .% : :b 0

0 .0 0.' 0 .0 0 O. ,

0

U 5 0 . .. .. . . .. :........ '........ . . . . .:............. ... "....... ...... •......................... ........ .. .

0 . ..

83 84 85 86 87 88 89 90 91 92 93 94 95 96 97Time (Year)

Figure 11-8. Price history of the cocaine market as defined by the street price index.

Cocaine Street Purity History10 0 . . . . . . ., . . .. . . . . . ..•, , , , , , , , , , , , . , , , , ,

9 0 . ............... ...... ........ ... ............ .......... ........ ......... ' ....... ........ .........................-

0. .o 0

S8 0 ... ...... ... ... ... .. ..... ... :--- o .........~~~~~~~0 ......... . . & . . . . . . . . o ............000 • 0

"['- : i o o : oy : "

2~~~ 70 ...... 0

o: 8 o 0.

01-2

"---0 ~ ~~ ~~~ 0o. : :o c7) 0 .. -.. - -0 ... ...... ... ....... :....... I ....... :...... .. " .. . ....... ....... ....... .. ....... I.... ... o . . ..

50 .. .

0I2

6 0 .. . . I .. . . . . . . . . . . . . .. .. . . . . . . . . . .

show a complete history of the cocaine market since 1983. With the exception of a few

prominent features, the plots show that cocaine prices have been in a general decline since

1983 while cocaine purity was on the rise prior to 1988 and has been in slow decline since.

It should be understood that the "control" parameter in this market is purity. When the

market supply is obstructed, the market compensates by lowering purity, which effectively

increases price. Although during shortages the gross unit prices do increase somewhat,

the bulk of the price-raising mechanism is through a lowered purity.

Figure 11-8 reveals that the price of cocaine, as reflected by the street price index,

can be characterized as a smoothly decreasing trend on which are superimposed a number

of distinct, but short duration, upward excursions or "bumps," which in all cases recover

to the original trend line. The following chapter presents evidence showing that

associated with these price excursions is a noticeable decrease in cocaine use. Chapter IV

discusses the probable causes for the price excursions.

11-21

CHAPTER III

THE RELATIONSHIP BETWEEN COCAINE PRICE AND USAGE

III. THE RELATIONSHIP BETWEEN

COCAINE PRICE AND USAGE

If one believes that the price excursions observed in the STRIDE-derived street

price index reflect meaningful changes in the price of cocaine as available on the streets of

the U.S., one would expect that these price excursions would cause some change in usage

patterns, since few would argue that the demand for cocaine has zero elasticity to price.This chapter examines several indirect measures that are logically related to cocaine use in

an attempt to confirm that the price excursions noted in the STRIDE-derived street priceindex do, indeed, reflect meaningful changes in the use of cocaine in the U.S.

There are two parameters relating to use: prevalence and total consumption.Prevalence addresses the number of persons using the drug, while consumption addresses

the total quantity of drug consumed by those users. These parameters are related throughthe shape of a differential distribution function of consumption, what may be termed the"usage distribution function" or, for brevity, the "user function." The user function, n(c),represents the number (n) of cocaine users that use c grams of cocaine per year in some

narrow range about c. Prevalence is defined as an integral of n(c) between any two limits

of c, and total consumption as the integral of c times n(c) between any two limits. The

shape of the user function as derived from the National Household Survey on Drug Abuse

(NHSDA 1992) data is approximately hyperbolic

n(c) a C

with sharp cut-offs in the vicinity of c = 0.1 gm/yr (one dose per year) and c = 500 gm/yr

(a reasonable physiological limit).

Because' of the hyperbolic behavior of the user function, prevalence is dominated

by the low-consumption end of the distribution and total consumption by the high end. In

general, any price-induced change in demand depends on where on the user function curve

the change operates. A large decrease in the prevalence of low consumption users haslittle effect on overall consumption, while a small change in heavy use prevalence has a

large effect on total consumption but little effect on total prevalence. Since the four

indirect usage indicators described below reflect different portions of the user curve, it

III-1

would not be surprising to find that the response to a price change may be different for the

four indicators described.

The indirect indicators of usage investigated are:

1. The number of emergency room admittances linked to cocaine at participating

hospitals in the Drug Abuse Warning Network (DAWN) data collection

program.

2. The positive test rate for cocaine in the Department of Justice Drug Usage

Forecasting (DUF) program. This program conducts semi-random drug

testing on several hundreds of arrestees in 23 major cities every calendar

quarter.

3. The positive test rate for cocaine in the testing program conducted by the

SmithKline Beecham Clinical Laboratories (SBCL). Typically, SBCL

conducts from 250,000 to more than 300,000 tests per month for a broad

spectrum of the American workplace.

4. The number of cocaine treatments delivered by treatment centers participatingin the Treatment Episode Data Set (TEDS) data collection program.

While none of these indicators is a direct measure of cocaine consumption (or prevalence),

each is logically linked to drug usage, and any price-demand elasticity in the market should

logically result in measurable effects in these indicators when price excursions occur.

To quantify the sensitivity of the indirect usage indicators to changes in price, a

measure of elasticity is defined to be the local non-dimensional slope of the indirect usage

indicator vs. price index curve. "Non-dimensional" implies a normalization by the

respective means of the two variables. The estimate of the elasticity of a usage indicator is

primarily determined by the periods in which largest excursions in the STRIDE street price

index and in the usage indicator occur. These periods typically extend for short periods of

time, during which large shifts in the underlying demand are unlikely. In the figures

presented in this chapter, the street price index is compared to each of the four indirect

indicators and in all cases the slopes are approximately constant. In all cases, the binning

parameter, L, is adjusted to match the sampling interval of the relevant indicator.

Figure III-I shows the correlation between the STRIDE street price index and the

number of DAWN cocaine emergency room mentions by month for ages 12 to 35. The

age range has been restricted to the younger population because the older population is

known to have a much higher probability of visiting the emergency room as the result of a

single use incident than the younger population. This fact can significantly bias the time

dependence of the probability that a drug use episode will result in an emergency room

111-2

visit. The top panel of Figure III-I shows the price -index, the middle panel the DA WN

emergency room numbers, and the lower panel the nondimensional data (generated by

dividing through by the means), along with the "best-estimate" regression line (bold line)

and the 95th percentile confidence interval (thin line) derived using a bootstrap procedure

on the data (Efron and Gong, 1983; Efron, 1990). Assuming equal errors in the two

variables, 10,000 bootstrapped replications of the data were used to yield a median valuefor the linear regression slope (used here as the "best estimate") of e = -0.63 + 0.06. The

assigned error corresponds to one standard deviation in the bootstrapped distribution.The 95th percentile confidence interval, as determined by the bootstrapped distribution, is-0.86 < e < -0.50, and the standard, least-squares regression coefficient is R = -0.71.

The broad 1990 feature is clearly resolved in both variables, but other features clearlyresolved in the price index are poorly resolved in the DAWN data. Nonetheless, the two

curves are not wildly discordant. Such a correlation was first reported by Hyatt andRhodes (1992, 1995); however, the claimed correlation was criticized as "coincidental,"i.e., "correlation does not imply causality," and several alternative interpretations have

been proposed for the DA WN data.

Figure III-2 shows the correlation between the STRIDE street price index and the

DUF cocaine testing positive rate. The top panel shows the price index, the middle panelthe DUF cocaine positive rate, and the lower panel the nondimensional data along with the

best estimate result using the bootstrapped linear regression procedure described above.

Assuming equal errors in the two variables, 10,000 bootstrapped replications of the datayield a best estimate of e = -0.29 + 0.07, a 95th percentile confidence interval of -0.51 <e < -0.19, and a standard, least-squares regression coefficient of R = -0.64.

Figure I1-3 shows the correlation between the SBCL cocaine positive test rate forthe combined workforce and the STRIDE-derived street price index. The top panel shows

the price index, the middle panel the SBCL positive rate, and the lower panel thenondimensional data along with the best estimate result using the bootstrapped linear

regression procedure described above. Assuming equal errors in the two variables, 10,000

bootstrapped replications of the data yield a best estimate of e = -0.60 + 0.13, a 95th

percentile confidence interval of -0.98 < e < -0.29, and a standard, least-squaresregression coefficient of R = -0.51.

Figure 111-4 shows the correlation between the STRIDE street price index and the

fraction of TEDS-reported treatments linked to cocaine (cocaine reported as primary orsecondary abuse drug). The top panel shows the price index, the middle panel the TEDS

fraction linked to cocaine, and the lower panel the nondimensional data along with the best

111-3

STRIDE Street Price Index

12 I . I . I ! 1

100

100 . . .. .. . . ............. .. . ............... ......... ............... ..-- - -- -.... . . .

C P 7 5 --- --- --. . . . . . . . -r. . . . . . . . .. . . a_. . . . . . . . . . . . . o . . . .

1988 1989 1990 1991 1992 1993 1994 1995

DAWN ER Corrielaetions g 23

9 1.......... ......... *...... .... . . . a........................ ........... .........

0 I

*0 .5........ . ...... ........................ a . a.. a% . . .. .0 -

0

6 . 1.. .. 5 .1 5 0. ... .. 75.

Reatv Price0

Figure~ 0l-I Cor0to ewe h TIEdrvdccieSre rc ne n hnube of DAW emrecomcoan etos

5 -- -- -- - .. .. .. ... .. .. .. .. .. .. .. . .. .. ...1.. ..

STRIDE Street Price Index125

CY' 5 . . . . . . . . . . . . . . . . . . . . . . .

000

40........................ .. . . . . .. . . . . . . . . .. . . . . .. . . . .

205I1987 1988 1989 1990 1991 1992 1993 1994 1995

Time (year)

DU CocaCorresatinPoniieRt

60

0

00

0.57 0.75198 19991 19 1.25 1.5 4 1.95

CrrelativePrc

Fiue1-.5CreainbtenteSRD-eiedcciesre rc ne n hD0 oan ostv ae

1. 5 .. ... .. .. ... .. ... .. .. .. .. .. ... ... .. .. .. ...1. .. ... .. .


50

E -

1992 1993 1994 1995 1996 1997

Time (year)

SmithKline Beecham Clinical Labs Cocaine Positive Rate1.30T

U 1.2 . . . . . ..---------...... ....- --- .... . ..... 0.............

0.8 0

0.7 0

a) 0

0 0

00

0)

0.7 r

0.52 1.75 199 1925 1995 19975

CrrelatiePice

Fiue11-.5CreainbtenteSRD-eiedcciesre rc ne nh

1.25in ..eecham... Clinical ...... Laboratories.... combined.. ..r.oce.ocin.poitvetes. rte

0 1-


0

100 : . . .. .. .. .. . . . . . . .. .. . .. . .. .. .. .. .. . . . . .. . . . . .. .. .. ... . . . . : . . . . .

E o .... .. . .... .. ....... L ... ...... . . . .0 .. 0 o 0 o o

050 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . ..-- --. . . . . . . . . . .. . . . . . . . . . ..

25I I I

1990 1991 1992 1993 1994 1995 1996 1997

"lime (year)

TEDS Cocaine to Total Admissions Ratio40

Co rr elaio

3 5 .... .... .. . ... -- - - - - -- -- - --- ----I- -= = = I I I

00

00

00

C

1990 1991 1992 1993 1994 1995 1996 1997Time (year)

Correlation1.5 . . . .

0 ~- - - - - - - -1. 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..

•- . ................ .. .. . -. ,-----o -o . . ............ o:................. -................

• a, . .0

4 0 .7 5 - - - - - - - - - - - - - - - -- - - - - - - - - - - - - -- --. .. .. .. . . . . . . . . . . , -. . . . . ..- - . .- -. .. .

0.5 1 . . . . . . . .

0.5 0.75 1 1.25 1.5 1.75

Relative Price

Figure 1114. Correlation between the STRIDE-derived cocaine street price index and thepercentage of TEDS treatment center cocaine admissions.

111-7

estimate result using the bootstrapped linear regression procedure described above.Assuming equal errors in the two variables, 10,000 bootstrapped replications of the datayield a best estimate of e =-0.38 + 0.05, a 95th percentile confidence interval of -0.49 <e < -0.29, and a standard, least-squares regression coefficient of R = -0.73.

Table III-1 summarizes the estimates of price elasticity as determined from the fourindirect measures of cocaine use. In interpreting the correlations observed between the

STRIDE price excursions and the four indirect usage indicators, it is appropriate torecognize that these usage indicators reflect completely independent processes fromdisparate populations. The fact that all four show substantial correlation with the

STRIDE-derived street price index strongly supports the hypotheses that the excursions inthe data do reflect meaningful changes in price and purity as experienced on the streets ofthe U.S., and that these excursions are indeed responsible for the changes in the usage

indicators.

The canonical value of the price elasticity of demand for cocaine that is commonlyused in the literature seems to be about -0.5, based on extrapolated knowledge from other

drugs such as alcohol and tobacco. Few measurements are available, however, and thoseare not in good agreement. Di Nardo (1993) finds no effect on cocaine use due to pricechanges, Grossman et al. (1996) find a range of-0.7 to -1.7, while Saffer and Chaloupka(1996) find a value of -0.28 based on an analysis of the National Household Survey on

DrugAbuse (see Saffer and Chaloupka, 1996 for a discussion and historical perspective).

Table I11-1. Summary of Price Elasticity Estimates for VariousIndicators of Cocaine Usage

Usage Best 95% Confidence RegressionIndicator Estimate Interval Coefficient

DAWN -0.63 -0.86 < e < -0.50 -0.71

DUF -0.29 -0.51 < e < -0.19 -0.64

SBCL -0.60 -0.98 < e < -0.29 -0.51

TEDS -0.38 -0.49 < e < -0.29 -0.73

While it is apparent that the "elasticities" of the indirect usage indicators developed

above are each related to the price elasticity of demand for cocaine, those relationships areimprecisely understood. In addition, as noted previously, the "elasticity" determined for

these indicators neglects intrinsic shifts in demand, believed to be small, that may haveoccurred during the period examined. Therefore, the slopes of the regression lines provide

111-8

only an approximate measure of the true price elasticity of demand. Nonetheless, the

"elasticities" determined above for the four indirect usage indicators are consistent with

the canonical value of -0.5 as being a reasonable estimate of the price elasticity of

demand, averaged over the entire user function.

111-9

CHAPTER IV

THE EFFECTS OF INTERDICTION CAMPAIGNS ON THECOCAINE MARKET

IV. THE EFFECTS OF INTERDICTION CAMPAIGNS

ON THE COCAINE MARKET

A. TIMING OF INTERDICTION EVENTS

This section examines the timing of major interdiction events, primarily in the

source-zone, as they relate to the price and purity excursions observed in Figures 11-8 and

11-9, and, based upon the timing coincidences between the price and purity excursions and

these interdiction events, argues that the interdiction campaigns are primarily responsible

for the excursions.

Figure 11-8 shows that, prior to 1989, the street price for pure cocaine was

generally falling. Undoubtedly, a number of factors contributed to this decreasing price,

e.g., the supply of cocaine increased markedly in the late 1970s and early 1980s from

increasing competition in response to the perceived growth in the U.S. cocaine market

throughout the late 1970s, only to be faced by a reduced demand as the national proclivity

for drug use began to wane in the mid-1980s.

After 1989, price and purity data show a less smooth history with several episodes

of rises and falls. One can postulate that the street price fluctuations seen in Figure 11-8

and the purity fluctuations seen in Figure 11-9 after 1989 are correlated with large-scale

attempts by the U.S. and source nations to disrupt the flow of cocaine into the U.S. The

four most prominent features in the data, those centered on 1984-85, 1990, 1992, and

1995, can be readily associated with the four major interdiction efforts spearheaded by the

U.S. and the source nation governments. These are (1) the DEA-led initiative known as

Operation ChemCon, (2) the initiation of the Bush Administration's War on Drugs, (3)

the conduct of Operation Support Justice III, and (4) the implementation of the shoot-

down policy against trafficker aircraft by the government of Peru.

Operation ChemCon was a major effort by the DEA to "bug" cocaine precursor

chemical containers at the source so that they could be tracked to processing centers. The

operation culminated in 1984 with the discovery of a very extensive laboratory complex

near the town of Tranquilandia, Colombia. Colombian police, DEA, and other U.S.

agencies seized enormous quantities of fuels and chemicals essential for the processing of

IV-I

coca paste into cocaine hydrochloride. The street price index and purity data indicate that

the repercussions of this operation on the cocaine market were immediate and significant.

Unfortunately, because cocaine market prices were in "free-fall" at the time, the resulting

price increases were small (but clearly resolved). The purity drop shown in Figure 11-9,

however, was deep and sustained for more than one year. Unfortunately, no analysis of

price and purity data was conducted at the time, and Operation ChemCon was deemed a

failure by the DEA. The industry's recovery, apparently completed some 18 months after

initiation, was presumably the result of adaptation. Although little, if any, quantifiable

evidence for this can be found, anecdotal accounts abound.

The major activity hailed as the War on Drugs, initiated by the Bush administration

in 1989, was a multi-faceted effort with a preponderant emphasis on interdiction. These

interdiction actions were primarily focused in the transit-zone of the Caribbean, but

enhanced operations in the source nations also disrupted the production/distribution

machinery of the cocaine industry. In 1989, most cocaine was transported by trafficker

aircraft to areas south of the U.S. border where it was smuggled across. In late 1989 an

extensive air surveillance system was put in place involving airborne early warning aircraft,

deployable ground-based and naval radars, and apprehension teams. As an example, the

number of airborne early warning aircraft flight hours increased nearly tenfold to many

thousands of flight hours per month at the peak in a trend nearly coincident with the risein price index. Sufficient operational activity was provided to cover completely three

principal air trafficking routes. The timing of the ramp-up of flight hours is consistent

with the hypothesis that the air surveillance system was a major (but not necessarily the

only) contribution to increasing the risk and cost to the air traffickers. Large dislocations

of coca workers were reported at that time in the news media. However, the interdiction

operation was not sustained because a large reduction in resources (the flight hours were

rapidly cut almost in half within a month) took place after September 1990 as U.S. military

forces deployed to the Persian Gulf in response to the Desert Shield build-up. This

drawdown of interdiction assets, along with trafficker adaptation, are likely causes of the

recovery of cocaine price and purity to pre-1989 levels, as shown in Figures 11-8 and 11-9.

The single largest adaptation effect was the establishment of new business relations, with

what are now known as the Mexican transportation cartels, for transportation services

over Mexican land routes.

Operation Support Justice III, initiated in late 1991, focused on suppressing the

movement of coca products out of Peru's Huallaga Valley. By early 1992, the Peruvian

Air Force aggressively pursued trafficker aircraft, resulting in a few, but significant, losses.

IV-2

This operation also was not sustained; it was concluded in May 1992 because of a

diplomatic incident. This action and the subsequent operation suggested that the

perception of risk was enough to raise the fees for transportation services from Peru to

Columbia, and that these increased fees propagated rapidly throughout the distribution

network to raise prices noticeably at all levels of the market.

By the spring of 1994, a major debate emerged over the legality of the U.S.

providing data to source-zone host nation forces on suspected drug trafficker aircraft,

when it could be reasonably anticipated that the host nations might shoot down suspect

aircraft failing to obey instructions to land. The refusal to provide such data culminated in

the suspension of all U.S. interdiction efforts within the source nations from May 1994

through the early part of 1995. In March 1995, nine months after the suspension of

interdiction efforts in the source nations relying on U.S. support, the price of cocaine in

the U.S. reached an all-time low.

The President proposed and the Congress enacted legislation in 1994 to enable the

U.S. to assist host nations with tracking data. In early 1995, the government of Peru

implemented a shoot-down policy against trafficker aircraft. Since April 1995, the

governments of Peru and Colombia, assisted by tracking data from many U.S. agencies,

have successfully shot down or otherwise destroyed dozens of trafficker aircraft flying the

so-called "air bridge" from the growing regions of Peru to the processing centers in

Colombia. As shown by Figure 11-8, from March of 1995 to September of 1995, cocaine

prices in the U.S. surged to a five-year high.

By October 1995, cocaine prices were again heading down (although the purity of

cocaine seemed not to have recovered fully). As of this writing (October 1996), the "air

bridge" has not been reconstituted and, once again, trafficker adaptation may be credited

for the price reversal. However, an alternative explanation has been suggested. Starting

in the middle of 1995, as cocaine prices were rising rapidly, several "kingpins" of the Cali

cartel were arrested or killed. Many Colombian growers and traffickers have expressed

the opinion (Weisman, 1995) that removal of dominance by the Cali cartel has opened new

avenues of competition, driving prices down. This point should not be overlooked.

Figure IV-1 is a reproduction of Figure 1I-8 annotated to show the beginning of

the eight major source-zone actions conducted since 1984, as identified by the historian

IV-3

Timing of Interdiction Events

1. OP. CH'ECON

S2. OP. BLAST FURNACE I,303. OP. BLAST FURNACE 11S30 0 .. ... """ .. .:. . . .: .... .. ... . . ....... ' ... . .: ....... I ...... ... .. .1..s':v "6 :hb •: • d hs " ...... .

eEs~ IUSh ON ORUGS; BEGINS5. OP. SVPPOR, JVS71E ,iI

0 1 6. OP. SUPPORT JOS7ETVCI 7 op SILVA VERDE (.CNP) 'C" 250 ... •°',, .. :; ... 2 ...... :....... ....... .......:...... ......... 7.. SP.O•,:_ 0 _ ' . ,.. ..... ~b!.......

250 00T DOWN POLICY IMPLEMENTED:CY) . : 8 ...

000) o

""0

2 0 1 0 . .. . . . . . . . . .. . . . . . . . . . . . .• o . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . .200

S150' 4 5 6' 7 '

,0 010 0 I I o

[1 O ... ... .... ... .........° .... .. ° .:° ..... ......".... I" ......

1 5 0 ....: ...: - :. . ... : . . . .:. .9. . . . . . . . . . . . .. . . . . . . . .. . . . . . . . . .(f~ 50

:6

0 0 : I

83 84 85 86 87 88 89 90 91 92 93 94 95 96 97

Time (Year)

Figure IV-l, Price history of the U.S. cocaine market with superimposed time markersshowing the timing of all major source-zone interdiction events since 1980.

of U. S. Southern Command.l In addition to the four events previously discussed, several

additional events are indicated on Figure TV-i1:

*Operation Blast Furnace/I. The U.S. disrupted the Chapare growing region ofBolivia using mobile teams of the Bolivian Army and police transported byU.S. helicopters.

* Operation Blast Furnace 11. Thifs was a follow-on disruption similar to BlastFurnace I.

*Operation Support Justice 1I7 Thifs operation, was characterized by much lessaggressive operations than Support Justice 0I and few end (arrest) actions. It

was terminated when the U.S. denied tracking data assistance from May 1994to December 1994 because of a dispute over Allied nation shoot-downpolicies.

*Operation Selva Verde. This operation was conducted by the ColombianNational Police against cocaine processing laboratory complexes. This event

The historian of U. S. Southern Command identified ten events - those shown on Figure -IV- plus OperationsSupport Justice (SJ) I and II. SJ I and II are not included in Figure IV-1 since further examination has shownthat these operations were not, in themselves, major operations in the source zone. Rather, they established theinternational cooperation, increased deployed assets, and provided training that was later employed effectively inoperations SJ III and IV.

IV-4

occurred during the period when the U.S. had terminated information sharingand support for Colombia and Peru because of legal restrictions on U.S.

' personnel.

Caution is warranted when associating cause with a single correlation, or even

with multiple correlations that occur without logical relationships. However, in view of

the consistent pattern observed between source-zone interdiction events and excursions in

the STRIDE-derived street price index, source-zone interdiction activities are the most

reasonable explanations for the major price and purity excursions. In fact, other than

random events, no other explanation for these consistent patterns has been offered. This

does not necessarily imply that all of the observable excursions in the price and purity data

are attributed exclusively to source-zone interdiction activities; however, in terms of

market economic dynamics, the mechanism is quite plausible and consistent with the data.

There is further evidence that the source-zone interdiction activities described

above did produce major market disruptions of the cocaine industry. Figure IV-2 depicts

Peru Cocaine Base Transport Estimates60

00

50.............. ... . ....... ........ ..... ............

-C0 00

E 40..... ............. ........... ...... . ............ ..... .................. ................

I.-300

0.0* 0- .. 9 .. . . . ... . . . ......... I . ...0 0*".c~ ~~~~~ 20 . .. . . . . . .. . . . .......... ...... . ................

0- 0-'2 _. ....... ... ..o . i ... .... ............ i............. G .... ................ .i ....° .. ... ...... ... .... ..P0

(A1 10 ..............

10.0

1991 1992 1993 1994 1995 1996 1997Time (Year)

Figure IV-2. Estimates of coca base transport on the air bridge from Peru to Colombia.

IV-5

the quantity of cocaine base transported along the air bridge as estimated by the Peru

Tactical Assessment Team2 (TAT). The figure shows a significant decline in cocaine base

transported by air from Peru to Colombia starting in the third quarter of 1991 until the

spring of 1992. Another significant decline began in March 1995, coincident with

aggressive interdiction by the Peruvian Air Force of trafficker aircraft that resulted in

shootdowns of aircraft resisting apprehension.

Classified radar tracking data confirm that air traffic all along the air bridge hasplummeted since March 1995 and the number of observed flights has declined to about

one-tenth of previous numbers. As of this writing, the air bridge has not been

reconstituted. It is now generally believed, and supported by intelligence and newspaper

accounts from the region, that trafficker organizations have been forced to much greater

reliance on river and land routes to export base materials out of Peru. Because surface

operations are much more costly and complex due to the high risks associated withmaintaining security, the industry may have undergone a structural and perhaps permanent

change with noticeably increased costs.

For severe disruptions of transportation of base from Peru to Colombia, the price

of base in the growing regions of Peru would be expected to drop dramatically because

the farmers cannot sell either leaf or base. Figure IV-3 displays data on the estimated costto produce base in Peru (open squares) and the price that traffickers pay for base in Peru

(open circles). The price data are primarily from the Peruvian Projecto Especial Alto

Huallaga (PEAH) (PEAH, 1996) and are based upon sampling of prices at 15 or more

locations in the Upper Huallaga Valley and a few other regions. The data prior to 1993,

which are less accurate, were compiled by the U.S. Agency for International Development

(USAID, 1981, 1983), from a variety of sources prior to the PEAH project. The United

Nations Drug Control Project also has collected leaf, paste, and base prices since 1989,

and these are consistent with the PEAH data (UNDCP, 1996.) The estimates of the costto produce base include the chemical, labor, and processing costs and PEAH prices for

raw coca leaf (or paste) (Cuanto, 1993). Sharp decreases in the price for base are

apparent and correspond to the timing of the War on Drugs, Operation Support Justice

III, and the implementation of the shootdown policy in 1995. This figure indicates that the

2 The TATs are multi-agency bodies that operate in each host nation embassy to support the activities of the DEAwithin the host nations. The TATs work closely with host nation police and military intelligence organizationsto monitor drug-related activity.

IV-6

Peruvian Coca-Base Price and Cost10 0 D . ,. . .o I . I . I I I . I I I I I

1000

" 7 5 0 ......... 6 -: .... ............ .... ...... . ..... . .o ......... ............. i..... ... ..... i.............00

U0

1989 19900 1 13 9 1 1

0 . .o . ...rC) 500 .... ...... ..... C . ......... .............. ...........

12 E : b

3_ 2 5 0 . .. . .................... ... . . . .. . ... . . ..... . . .

1989 1990 1991 1992 1993 1994 1995 1996 1997

Time (Year)

Figure IV-3. History of coca base price (circles) and estimated cost (squares) in Peru.

average price of base plummeted to or below the average cost to produce during 1995,

and that the margins between cost and price remain far smaller than those throughout the

1991-1995 period. These estimates of reduced profitability since the shootdown policy in

1995 are consistent with reports during 1995 of the displacement of large numbers of

cocaine workers in Peru (Craig, 1995), and commensurate requests for food assistance for

displaced workers (U.S. Embassy, Peru, 1995).

B. THE LONG-TERM BENEFITS OF INTERDICTION

In the preceding section, we argued that several interdiction actions, primarily in

the source-zone, have been responsible for short-term, but significant, increases in the

street price of cocaine in the U.S. These price increases, in turn, have been associated

with decreases in cocaine use patterns. We cannot, with confidence, make definite

predictions about the long-run price effects of interdiction. Interdiction is a recent policy,

and even during its brief life span, it has not been consistently pursued. We do not know

what long-term market responses interdiction will produce, and no amount of study, at the

present time, can answer this question.

We nonetheless challenge the common view that interdiction necessarily is

ineffective in the long run and argue instead that properly focused interdiction efforts can

IV-7

force a structural change in the market, resulting in a long-term increase in price. The

logic underlying this argument is twofold. First, cocaine production and distribution are

motivated by profit and attempt to establish the most efficient distribution channels. If a

well-conceived interdiction activity permanently blocks an established distribution channel,

the enterprise is forced to adapt by developing an alternative and less efficient channel,

with commensurate increases in costs. This permanent increase in costs will be reflectedin a permanent increase in prices. Second, in a market in which prices are declining

because of increasing competition and decreasing demand, the sudden appearance of newrisk within the industry can cause a sudden change in the rate at which new competition

enters and correspondingly the rate at which prices decrease.

As an example of the first mechanism, we have already mentioned that a majorresult of the 1989 "war" was the establishment of new business agreements between the

cocaine producing organization and the recently formed Mexican transportation cartels.

The business sanctions imposed by this new partnership were effectively an immediate

doubling of the marginal costs of producing and importing a unit of product, since the fee

charged by the Mexican cartels is typically one-half the cocaine to be transported.

Many studies also assume that inventory accumulation blunts the price effects ofinterdiction. This claim is no more than partially correct. First, many interdiction effects

cannot be blunted by inventory accumulation. The Peruvian air bridge, for instance,

restricts the supply of coca base to Colombian laboratories, but also raises the costs of

accumulating inventories. In this respect, the air bridge policy should act to lower

inventories, not raise them. Some interdiction policies, such as coca leaf eradication

programs, may be blunted by inventory responses, but many other forms of interdiction,

such as raising transportation costs, make inventory responses more costly as well.

Second, long-run adjustments, such as those that might occur in inventories, donot imply that interdiction has no effect on the street price of cocaine. The mere

expectation of interdiction can induce higher inventories, thereby raising the marginal cost

of production, and thus the street price of cocaine. Price spikes may disappear, but theprice of cocaine will be permanently higher, due to higher inventory costs and other

adjustment costs. Over time we should expect the price spike to diminish, and the price

effects of interdiction to be spread out and smoothed. But the price effects will remainreal, even though they become apparently invisible; they simply have become permanent

and smoothed, rather than spiked and concentrated at particular points in time.

IV-8

In the long run, if interdiction is pursued regularly, the price series for cocaine may

be perfectly smooth. In the relevant graphs, we would see that stepped-up interdiction has

no effect on market price at the street level. Such results, however, would not imply that

interdiction is ineffective in raising price. Interdiction still would be raising the overall

base for (temporally flat) cocaine prices by raising inventory and adjustment costs.

Observational data suggest that a structural change occurred in the cocaine

industry in 1989. Figure IV-4 shows a plot of the median gross (i.e., uncorrected for

purity) unit-price for all STRIDE data since 1983. Each open circle represents the median

value of 100 time-sequential purchases, and the bold line is the usual nonparametric

regression curve. The salient feature of the curve that supports the argument for a

structural change in the industry is the sudden change in slope or "kink" that occurred

early in 1989. In view of the temporal correspondence of this "kink" with theimplementation of the War on Drugs, we attribute this structural change to the imposition

of the extensive interdiction network placed between the product source and its market.

Time Sequence of STRIDE Gross Purchase Price150

000

N00~)

0' 00

0~0C'o

.. .. .. .. .: . o .. .100- .... ... O.. .o. ............... ............................... ......... ........ ......... ...

10:

a-

• . d 2 . 0 .0...

Fiur •. o o I d . uipc f50 ............... . ....... ý . ..

IIV-9-

0- ... ' . ..

Fgr IV4 Tim hitr of STID median grs puchs untprc fr o.•. om

198 tomdi9

Accepin th hyohei thtteitrito ou;o h a nDuscue

profigure author oftime effects of STRDEmerous othr fors onurchaine untprices inrom eal

IV-9

Indeed, one can readily argue that U.S. Customs Service actions to inhibit import of

cocaine at the U.S. borders, aggressive domestic law enforcement, and severe sentencing

of drug dealers are all contributing components of policies aimed at reducing supply.

However, we are unable to identify specific dramatic changes in these aspects of supply

control that correlate in time with the observed shift in the price data slope.

The dramatic change in the slope of the street price of cocaine occurring in 1989 is

also clearly evident in the normalized price of cocaine - the street price index. Figure IV-5

shows the street price index from 1985 along with the "trend" line that characterized the

price between 1985 and 1989. The trend line indicates that between 1985 and 1989, the

price index was declining, following an approximately exponential path.3 The smooth thin

line is the result of an exponential fit to the data between 1985 and 1989. The

extrapolated street price index will serve as the basis for our estimate of the cost-

effectiveness of total interdiction in section C.2. below.

Figure IV-6 presents a more detailed depiction of the last eight years of the price

index shown in Figure IV-5, highlighting that several transient excursions have occurred

since 1989 with varying effects. The large 1990 feature, we have hypothesized, was due

to the combined effects of all interdiction efforts - source-zone and transit-zone - but the

other features (after 1992) are primarily the result of source-zone activity. We have

previously noted that, when the U.S. "turned off" its source-zone interdiction machinery in

late 1994, the street price index reached an all-time low of about $50 to $55 per gram

within nine months. This is consistent with the existence of a baseline "floor" value for the

market on top of which all transient increases are superposed. This baseline, shown as the

solid horizontal line in Figure IV-6, at a level of approximately $55 per gram is interpreted

as the baseline price index in the absence of source-zone interdiction efforts.

When averaged over the last five years, the price index accounts for approximately

a 30 percent increase over the floor baseline. This average effect is shown as a dash line in

Figure IV-6. The difference between the floor price and the average level is interpreted as

the average benefit of the source-zone interdiction program since 1992.

3 This functional behavior is the subject of continuing research to be addressed in a forthcoming IDA paper.

IV- 10

U. S. Cocaine Market Price History300. . . . . . . .

0 .

0 25....... ........................... ............................E 2 0 .. .. . .. .. .

0 0.

0)

0 . 0

X

C

00q) 100 .0

0~0

0 0 0: a o0 0-*-- 50

011

85 86 87 88 89 90 91 92 93 94 95 96 97Time (Year)

Figure IV-S. Street price index of cocaine since 1985.

U. S. Cocaine Market Price History20 .C . . . . .

-E 0

a)j 0a)

0 0:I

89 .90 90 929049 69100 ~ ~ ~ ~ ~ ~ T m (Y ear). ý ................................

Fiur IV 0 Dealdve 0ftesre rc ne ic 99

0 0 0 0 0v0-il

Additional support for the thesis that source-zone interdiction activities can cause

structural changes in the U.S. market is provided by Figure IV-7. This figure shows the

STRIDE median normalized unit-price and median purity for all transactions restricted to

small purchase quantities (in the normalized volume range of 0 to 10 grams). This is the

volume range typically used as representative of the "retail" street market. Evident from

this figure is a distinct decrease in median purity, driving a commensurate increase in price,

beginning in April 1995. The two straight lines in each figure are meant to point out the

trend baselines prior to and after the transition time of April 1995. This transition time

correlates well with the initiation of the shoot-down policy by Peru. This coincidence is

consistent with the thesis that the shoot-down policy is responsible for the observed price

and purity changes. One cannot determine if these higher "retail" prices will continue

indefinitely, or, alternatively, if effective adaptation by traffickers will eventually return

prices and purity to the pre-April 1995 levels. Nonetheless, the effect is clearly of

significant duration and continues to persist to date.

Cocaine Retail Price History200 1 1v !,

00 0 (0 . o o d o 0

0 0o o' 00) 0 00 00 0, 0 0 0 00 0 0 000 (b 0- ' -0 00 ('.00%:00o00 O 0%:0 C9 0

0 O00 . 0 oV o0 0oi

0 50-O 50 ..................... ..................... ............ ....... .................. .... ....................

•0

0 , , I , , , , , , I , ,

92 93 94 95 96 97

Time (Year)

Cocaine Retail Purity Historyloo I

0009 0 . ... ............ ..... ................ ................... .. ...... .......... ...... .I............ .. .. ...." 0 o o 0 o o : oo 000 0 080 ." 9oo °00 > o .".o L9. 0~o: .. ... nbO~o .o . o, 0" 00o 0o . . . . ......... " .... 0 ....... ..

S:o o o :% o. n ° o :-0 Oob •0 0%•0 . 0 .. 0 °

0 0 0 0 0

Ss .... ..................... i............. i........... N . .... 0 ........ _............ .. * , , ;0q , -.0 , , 00

400 0

92 93 94 95 96 97

Time (Year)

Figure IV-7. History of median normalized unit-price and median purity for all STRIDEpurchases in the range (0 and 10 grams) of normalized purchase volume since 1992.

IV-12

Earlier in this chapter, we associated the shoot-down policy with markedly

decreased traffic along the air bridge and with a significant decrease in the profitability of

producing base in Peru. These occurrences, considered collectively, lend strong support

to the thesis that this source-zone effort has indeed caused a structural change in the

production and transportation elements of the cocaine industry, resulting in higher long-

term prices of cocaine in the U.S. domestic market.

C. THE COST-EFFECTIVENESS OF INTERDICTION

The preceding section argued not only that interdiction activities have produced

transient increases in the street price of cocaine, but also that these activities have resulted

in long-term price increases and a commensurate decrease in the consumption of cocaine

in the U.S. This section develops rough estimates of the cost-effectiveness of these

interdiction activities. While the estimates provided are necessarily imprecise, they are of

the correct order of magnitude and are, therefore, useful in assessing the relative cost-

effectiveness of interdiction efforts within the context of the overall program to reduce

cocaine use in the U.S.

1. Cost-Effectiveness of Source-Zone Interdiction

The previous discussion of Figure IV-6 suggested that the transient price increases

associated with source-zone interdiction activities since 1992 have caused the street priceof cocaine to be, on average, approximately 30 percent greater than the floor baseline

price. During this time, the U.S. is estimated to have spent approximately $200-$400

million per year4 for source-zone interdiction (ONDCP, 1996), i.e., interdiction against the

production or transportation of base or precursor chemicals. If a nominal price-demand

elasticity of -0.5 is assumed, then the 30 percent increase in price translates to a 15

percent reduction in demand. If cost-effectiveness is defined as the number of dollars

needed to achieve a one percent decrease in cocaine demand, then a rough estimate for the

cost-effectiveness of source-zone interdiction is:15% Reduction in Demand

Cost-Effectiveness = = 0.08-0.04$200- $400 million $ million

or approximately $13 - $25 million per year to effect a one percent reduction in demand.

4 The estimate of $200 to $400 million as the average annual cost over the 1992 to 1996 period includes thefunding for "International DEA," "International Narcotics Matters," and the source-zone portion of "Departmentof Defense Interdiction" and excludes funding for "Agency for International Development" (ONDCP, 1996).

IV-13

This empirical estimate is clearly directly applicable only to the experience of 1992

to 1996; i.e., as in any socio-economic system, past performance is no gauge for future

performance. In addition, caution is warranted in extrapolating any such estimate of

marginal cost-effectiveness to large changes, because the cost-effectiveness is likely tovary in a highly nonlinear and chaotic fashion as the conditions move far from the original

equilibrium point. Nonetheless, this admittedly crude estimate does suggest that a well-conceived source-zone interdiction strategy aimed at denying production andtransportation to the cocaine industry can be relatively cost-effective.

2. Cost-Effectiveness of Total Interdiction

In the previous section, we have argued that the transitory excursions in the

cocaine street price index can be causally linked to source-zone interdiction activity. We

now proceed to more speculative ground and argue the case that the overall effect of

interdiction has been to keep prices elevated well above what they would have been in theabsence of such interdiction.

As noted previously, Figure IV-5 indicates that between 1985 and 1989 the streetprice index decline is consistent with a smooth, approximately exponentially decaying

trend. In mid-1989, this trend line was broken drastically by a sudden reversal of pricesthat reached a peak in 1990 with a doubling of prices. Although prices recovered by1992, they never returned to the pre-1989 trend line. Instead, the price index after 1989,disregarding the numerous transient fluctuations, adjusted to a new trend line with a much

shallower slope. We believe that this transition was not spontaneous but was induced bythe introduction, via the War on Drugs, of an extensive interdiction machinery placed

between the source of the product and its marketplace.

In the absence of the War on Drugs, the accepted economic theory of

monopolistic competition argues that prices would have continued to decrease, eventually

converging to a value equal to the long-run average cost. It is generally understood that

the costs of producing and transporting cocaine to the U.S. are small, i.e., currently on the

order of $5 per gram. However, while it is clear that the street price index remains well

above the long-run equilibrium price, there is no accepted model with which one might

project the path that the street price index would have followed in the absence of the Waron Drugs, or the number of years that would pass, in reaching the equilibrium value.

As a rough approximation, however, we believe it is reasonable to hypothesizethat, in the absence of the War on Drugs, the street price index would have continued to

IV-14

follow the exponential trend line observed until 1989, asymptotically reaching an eventual

price of approximately $35 per gram. One could readily argue that this value exceeds the

"long-run average cost," but the intent is to develop a rough, but not overstated, measure

of the impact of the War on Drugs. Note that this assumed price decline to $35 per gramincorporates the influence - which we believe to be substantive - of border protection, law

enforcement, and legislative actions that were initiated in the 1970s and increased

significantly throughout the 1980s.

Following the Bush Administration's war on drugs, the street price and purity of

cocaine have never again returned to the pre-1989 path. (Note: The disruption of the air

bridge in 1995 has shown a similar kind of sustained effect, but on a smaller scale.)Instead, the street price index since 1989 has been kept elevated at or above a "floor"

price of approximately $55 per gram. Consequently, we argue that the street price indexincrease above the extrapolated value of $35 per gram to the $55 per gram floor price is

an overall measure of the efficacy of the entire interdiction effort in the transit and sourcezones, i.e., the integrated result of U.S. and transit and source-zone government efforts to

disrupt market infrastructure and increase production and transportation costs.

Assuming a nominal value of-0.5 for the price elasticity of cocaine demand, it is

possible to translate the approximate 60 percent increased price resulting from the overallinterdiction effort into a 30 percent reduction in demand. Since we spend approximately a

range of $1 billion to $2 billion annually (prior to 1993, $2 billion annually) on this

interdiction effort (ONDCP, 1996), the cost-effectiveness can be crudely estimated as

30% Reduction in Demand %Cost-Effectiveness = =0.03 -0.015$1,000 - $2,000 million $ million

or approximately $30 million to $60 million per year to effect a one percent decrease in

demand.

Admittedly, the above derivation of an estimate of the cost-effectiveness of overall

interdiction is imprecise and subject to reasonable question. Nonetheless, we believe that

the data support an approximation of the cost-effectiveness of interdiction in the range of

tens of millions of dollars per one percent decrease in consumption.

IV-15

CHAPTER V

RECONCILING EMPIRICALLY DERIVED ESTIMATES OFCOST-EFFECTIVENESS WITH PREVIOUSLY PUBLISHED

ESTIMATES

V. RECONCILING EMPIRICALLY DERIVED ESTIMATES OFCOST-EFFECTIVENESS WITH PREVIOUSLY PUBLISHED

ESTIMATES

The estimate of the cost-effectiveness of source-zone interdiction (and the less

firmly based estimate of the cost-effectiveness of interdiction overall) presented in thepreceding chapter stands in contrast to a previously published and widely distributed

estimate derived through the use of a model to describe cocaine supply and demand. In a

paper titled Controlling Cocaine (Rydell and Everingham, 1994; see also: Rydell et al.,1996), analytical measures were developed to compare the cost-effectiveness of several

supply control programs versus the cost-effectiveness of selected demand controlprograms, notably the treatment of cocaine users. That study, using modeling as its

primary methodology, concludes that the cost of reducing cocaine consumption by one

percent via source-country controlI would be $783 million per year as compared to a costof only $34 million per year to accomplish the same reduction via treatment programs.The $783 million figure is 30 to 60 times larger than the cost-effectiveness values of $13

million to $25 million obtained in Chapter IV of this paper.

The measure of effectiveness employed by Controlling Cocaine is a reduction in

cocaine consumption over a 15-year evaluation period that is equivalent to a reduction of

one percent in the present year. This distinction is significant since interdiction and

treatment act on different time scales. Interdiction activities produce the bulk of their

effects in the first year with perceived or actual supply shortages driving up prices andreducing consumption. Without continuing interdiction efforts, the reduction in

consumption levels would reverse in time. On the other hand, treatment programsproduce only a small portion of their overall effect in the first year (20 percent as reported

in Controlling Cocaine). The major effect of treatment is the cumulative effect over the

future years of permanently removing users.

In developing the above cost-effectiveness estimates, it was recognized that a wide

disparity existed between the empirically derived value and that of the earlier work.

Controlling Cocaine uses the term "source-country control." It appears to encompass those activities that have

been referred to in this paper as "source-zone interdiction." In this chapter, the terms are used synonymously.

V-1

Although the empirical estimate of source-zone cost-effectiveness is admittedly imprecise,

it is based on consistent data and sound logic; hence, it can be expected to be accurate in

order-of-magnitude if not in detail. Consequently, some attempt to discern assumptions ormethodologies employed in Controlling Cocaine that might explain the large discrepancy

between the two findings is warranted.

The examination here has necessarily been limited in scope in keeping with the

restricted counterdrug role of the Department of Defense. The review has primarily

focused on key aspects related to the extremes of the findings, i.e., that treatment is 23

times more cost-effective than source-country control in decreasing consumption, which

contrasts markedly with this paper's conclusion that a well conceived source-zone

interdiction program can equal or exceed the cost-effectiveness of treatment.

Five aspects of the assumptions and methodology employed in Controlling

Cocaine appear to account for the major differences in conclusions. These are:

1. Use of an inadequate and incomplete model for cocaine supply-controlprograms that:

a. Relies on the quantity of product seized and the number of personsarrested as the cost-driving factors, to the exclusion of more importantfactors.

b. Assumes weak, nearly linearized, functional dependencies to relate causesand effects, in contradiction to the observed highly nonlinear effects thatare consistent with basic economic principles.

2. Assumption of an additive cocaine market pricing structure that is inconsistentwith the available data.

3. Use of erroneous results from a demand model as the primary evaluation toolto estimate changes in future cocaine use.

4. Extrapolation of the marginal cost of additional new treatments that isinconsistent with actual treatment experience.

5. Inclusion of inappropriate costs in determining the costs of source-countrycontrol and interdiction.

These factors contribute to substantial structural problems with the modeling approach

used in earlier research, and are discussed in turn in the rest of this chapter.

A. INADEQUATE AND INCOMPLETE MODEL

The cocaine supply and supply-control models developed in Controlling Cocaine

focused on seizures as the driver of source country control effectiveness. While the model

V-2

acknowledges that source country control activities may also produce some indirect

increases in financial sanctions and processing costs as seizures increase, the authors

contend that these indirect costs constitute a "second-order effect." With this fundamental

assumption - that source country interdiction activities can achieve an effect directly orindirectly only through the quantity of coca (or cocaine) seized - it is not surprising that

the modeling approach used in Controlling Cocaine should find such interdiction activities

to be cost-ineffective. The value of the product seized and the costs of replacement aremarkedly reduced as the distance from the growing areas is reduced; i.,e., seizing coca leafin Peru or Bolivia is not likely to be an effective strategy since coca leaf typically accounts

for only a very small portion of the cost of processed cocaine. Since cocaine and cocaine-precursor commodities have little value in the source countries, "routine" seizures of

cocaine or cocaine-related materials in these countries have little effect on the market.Routine source zone seizures or loss of physical assets are viewed by the cocaine industry

as little more than a nuisance cost of doing business.

Controlling Cocaine modeled the quantity seized as being proportional to theamount of funds expended on source-country control. This is a questionable assumption,as it assumes that all expenditures are of equal value, neglecting the differing effectiveness

of alternative strategies, and incorrectly assuming that the objective of source zone

expenditures is seizures. The U.S. interdiction strategy in the growing nations is notaimed primarily at seizing product but at significantly disrupting and increasing the costs

of production and distribution.

The supply-control model constructed for Controlling Cocaine assumes that all

costs to producers and to the agencies attempting to control supply are directly related to

the quantity of product seized and/or the number of arrests, and, in all cases, therelationship is assumed to be a weak, nearly linear, function. These assumptions are

inconsistent with observed events. For example, Controlling Cocaine assumes the cost ofprocessing cocaine to be proportional to the quantity of product seized raised to the 0.44

power. This may or may not be correct, but it is irrelevant, since highly nonlinearincreases in processing costs can be, and have been, created without significant product

seizures. One dramatic example is the DEA raids on the Tranquilandia laboratory

complex, which raised the cost of processing chemicals enormously, causing greatlyincreased cocaine processing costs. During this time, little or no product was seized.

Similar concerns are noted with the modeling of "financial sanctions" as linear

functions of product seizures and the number of arrests, in detail (again these may or may

not be accurate), while failing to include transshipment costs. In the previous chapter, we

V-3

pointed out a major example of this mechanism at work by arguing that the direct result of

the 1989 war on drugs was the appearance of the Mexican transportation cartels with their

associated costs, which increased costs to traffickers enormously within a period of one

year during 1989 while seizure of product declined.

A second example of nonlinear effects on transshipment costs is the more recentmajor disruption of the air transportation network within the growing regions. The cocagrowing areas of Peru and Bolivia are located in isolated regions with limited to

nonexistent established transportation networks. The very remoteness of these areas hasprovided a major obstacle to host nation efforts to redirect farming to other more socially

desirable crops, since such crops are uneconomical for air transport. Coca base has

historically been moved from these remote areas to Colombia for processing almost

exclusively by small aircraft capable of operating from readily established, small dirt

airstrips. This air-based transportation system has become known as the "air bridge."Source zone interdiction strategy has increasingly been aimed at, and has been most

effective against, this air bridge.

The recent aggressive actions of the Peruvian Air Force vastly increased thelikelihood that a pilot illegally traversing the air bridge would encounter extremely serious

consequences. Not surprisingly, such directed efforts drastically increased - by orders-of-magnitude - the "going rate" charged by pilots to transport coca from Peru to Colombia.

Carried to the extreme, a successful source zone strategy to interdict the air bridge wouldtotally stop all flights by making the risk unacceptable to pilots. While such an interdictionprogram would be an overwhelming success, it would be "modeled" as totally ineffective

as measured by quantity or assets seized since in such a case the quantity of seizures (orassets or arrests) would be zero. Since Controlling Cocaine modeled seizures asproportional to source zone expenditures, that model fails to capture the dramatic effects

of Peru's shoot-down policy, since no large increase in expenditures is associated with this

change in strategy.

Disrupting the entire transportation system isolates production from its

marketplace causing huge inventory back-ups and lowering the value of the product at the

source. As a consequence of the air bridge attack of 1995, disruption of air transportation

caused the entire coca base production system to operate with significant losses (Figure

IV-3) for up to eight months. Farmers growing coca with the expectation of receiving$600/kg for base were forced to sell for only $200/kg - a distinct loss to them. Such

nonlinear events are excluded from the model central to Controlling Cocaine, but they

V-4

represent the dominant effects of interdiction, in contrast to the direct losses of seized

quantities, equipment losses, and arrests.

Such nonlinear effects are the rule in this economic environment, not the

exception. The failure of the Controlling Cocaine model to include these dominant

nonlinear effects makes the use of such a model problematic and incomplete as an

analytical tool.

B. ASSUMPTION OF AN ADDITIVE PRICING STRUCTURE

Appendix A of Controlling Cocaine provides the equations used in that study for

determining the flow of cocaine product and the establishment of associated prices

through a variety of postulated stages in the production/distribution network. As stated in

that report:

The total cost to producers at a given production stage consists of thepurchase cost of the net product from the previous stage which providesthe input to the current stage..., plus the processing cost of converting thatinput to gross output, including all capital, labor, and material costs..., plusthe cost to producers of supply-control financial sanctions, minus theoffsetting revenue from consumption of this stage's product....

The price of the net product from a production stage is the total costdivided by the net product. This equation uses our assumption that eachstage of the cocaine supply process is a competitive market.

Implicit in these formulations is the assumption of an additive pricing model, i.e., if

a cost is added at an early stage in the supply chain, the absolute value of that cost is

added to the follow-on stages. The marginal cost of production for one gram of cocaine is

very small, e.g., approximately $1 to $2 per gram, with transportation adding perhaps a

few dollars per gram. Hence, under the assumption of an additive model, an increase of

significance as measured by a percent increase in the costs of early stages in the supply

network would have very little effect on the eventual retail price in the U.S., where a gram

of cocaine sells in excess of $100.

As discussed at length in Chapter II, the assumption of an additive pricing

structure is in conflict with the data on cocaine prices derived from the STRIDE database.

Instead, the pricing structure appears to be multiplicative in nature. The assumption of an

additive pricing structure, prima facie, strongly reduces the likelihood that any action

V-5

conducted early in the supply/distribution network will be modeled as having substantive

impact on prices to cocaine consumers in the U.S.

C. UNRELIABLE ESTIMATES OF PREVALENCE DERIVED FROM THEDEMAND MODEL

Controlling Cocaine used a model of cocaine demand that divides users into two

groups, heavy users and light users, based on consumption. These definitions are not

identical with, but are close to, those of the National Household Survey on Drug Abuse

(NHSDA) "casual users" and "monthly users." Controlling Cocaine employs this model

as the basis for evaluating, and thus comparing, the manner in which the supply and

demand control programs affect the demand for cocaine into the future. The methodology

involves a numerical fit to past prevalence data on cocaine use and related cohort retention

rate to determine four free parameters of the model. With these four parameters

determined, the resulting model was then used to project future use. The values of the

four free model parameters used in Controlling Cocaine project a flat or increasing trend

of cocaine consumption over the period from 1992 to 2000+. This section argues that

these parameters were improperly obtained and that a proper determination of these

parameters, using the same prevalence data (but ignoring the cohort retention rate) as

used by Controlling Cocaine, leads to a forecast for future cocaine consumption that is

declining. More recent NHSDA data on estimated prevalence are presented that confirm

that the trend is decreasing.

The specific demand model used in Controlling Cocaine is described in Modeling

the Demand for Cocaine (Everingham and Rydell, 1994; see also: Everingham et al.,

1995). This model assumes that the number of light and heavy users during any year can

be represented by a pair of coupled equations with four free parameters that are assumed

to be constants over the entire range of years for which the model is applied. This model

is used to evaluate usage at yearly intervals, which can then be compared to actual data to

fix the values of the four free parameters.

In order to estimate the four parameters of the model, Modeling the Demand for

Cocaine constrained their allowable range to conform to judgments regarding the

reasonableness of such values. There are several "reasonable" but nonetheless

questionable assumptions involved in the model and in the constraints placed on the range

of parameter values. These are:

1. For simplicity, the model assumes that the parameters are constant over all ofthe years of interest, despite recognized changes in the popular acceptance of

V-6

drugs and in potentially relevant economic conditions over the period ofinterest.2 Another unlikely consequence of this assumption is that modeledprobabilities for a user leaving any category of usage are independent of howlong the user has been in that category.

2. The model assumes that light users progress to become heavy users as apercentage of the number of light users. One might suggest instead thatduring a time of increasing fascination with drugs, the rate at which light usersbecome heavy users might be driven alternatively (or additionally) to someextent by the number of heavy users. The model restricts this possibility byrequiring that all parameters be positive.

3. The model assumes that, over a year's step, no individuals progress from thenon-user pool into the heavy user pool.

To find the "best-fit" values of the four parameters, the authors of Modeling theDemand for Cocaine employed a trial-and-error search process to arrive at a subjective"good fit." The "good fit" was required to:

1. Produce a "near-minimal" mean square error between the modeled totalprevalence and the available prevalence estimates for each of 10 survey yearsover the period 1972 to 1991.

2. Reproduce the trend and approximately match the fraction of heavy users forthe three years (1985, 1988, and 1990) for which data were available.

3. Provide a model value for the "ten-year cohort retention rate" that was closeto the "observed value" (actually an average of three reported values).

The rationale for item 3 above, which constitutes an additional constraint on the

values of the four coefficients, is unclear. The cohort retention rate cannot be determined

from the data presented in Modeling the Demand for Cocaine, other than in the average,and the resulting retention rate curves are inconsistent for the three base years of the

calculations. It appears that the results are very sensitive to the application of any cohort

retention rate constraints, particularly the year chosen to best match with the "observed

value." Modeling the Demand for Cocaine compromised a good statistical fit to theprevalence data in order to "force" a fit to a questionable retention rate. The very fact that

completely different conclusions arise when the last "requirement" is added proves that the

assumed underlying model is wrong and/or that some or all of the data are suspect.

2 Indeed, the cohort retention rate data utilized by Modeling the Demand for Cocaine, for threeseparate years, exhibits strong temporal tendencies in direct contradiction to the assumption of staticparameter values.

V-7

Figure V- I displays the fit obtained by Modeling the Demand for Cocaine and that

obtained in this work using a formal least-squares fitting procedure that estimated the

same four parameters in the model of Modeling the Demand for Cocaine, but without the

cohort retention rate constraint. Both procedures used the identical prevalence and

incidence data for the fit.

Total Number of Users15. . . . . . . . .

10 - - - - - -. . . . .

En

0 5o ...................! .................- --. ......... ... i .............

01960 1965 1970 1975 1980 1985 1990 1995 2000

Time (year)

Number of Heavy Users2.5 . . . . .

2 - - - - - - - - - -. . . . . . . . . .. . . . .. . . . . .,. . . . . . . . . . •.. . . . . .:.. . . . . . . . .- - - - . . . .. . . . . . . . . . . . . .

.0.

0.5

01960 1965 1970 1975 1980 1985 1990 1995 2000

Time (year)

Figure V-1. Result of two fits to the same data for cocaine prevalence (solid symbols).Open squares are the "best fit" results as determined by Everingham and Rydell (1994).

The solid line is the "best fit" result of this work.

V-8

The figure shows the resultant "best fit" to the total number of cocaine users and

the number of heavy users as determined in Modeling the Demand for Cocaine (open

squares) and the "best-fit" determined in this work (solid line). The solid circles represent

the prevalence estimates as determined from the NHSDA. The solid diamond symbols

represent recent data that were not available at the time of the original study. These data,

which were not considered in either of the fitting procedures, are included to show the

predictive ability of the two fits. The new data points for total and heavy use prevalence

are those from the NHSDA "past 12 months" and "monthly users" categories scaled to

join smoothly with the original data since the NHSDA categories are not exactly the same

as those of the original study as defined by Everingham and Rydell (1994.)

Although the formal fit is greatly improved by removal of the cohort retention rate

constraint, this fit yields values for the four parameters that no longer seem plausible when

these parameters are interpreted as transition probabilities from one state of use to

another. A great deal of discussion on the significance of these model parameters and the

requirement to use cohort retention rate as a constraint for the model is possible, but is

neither warranted nor productive. Clearly, Controlling Cocaine's model (as constrained)

fails to describe the trends in the data accurately, especially in the future years. Figure V-1

shows that the "best-fit" to the data as reported in Modeling the Demand for Cocaine

underestimates (by approximately 30 percent) the total number of cocaine users over the

years near the peak of the epidemic (1979 - 1985) and overestimates the number after

1992. Their model is a simple, heuristic one with little or no significance to the transition

rates that are used as fitting coefficients; its primary value is in reproducing the functional

form of the data. With the removal of the cohort retention rate as a constraint, the model

does an excellent job of fitting the original data and predicting "future" trends. This flawed

model is still in use today (Rydell, 1996) and is the basis for predicting increasing cocaine

use in cocaine control program analyses, despite the fact that survey data since 1992

actually show declining cocaine use.

Since Controlling Cocaine and subsequent publications have used this model to

determine demand over its 15-year evaluation period, its estimates of cost-effectiveness

are affected by the specific values assigned to the four parameters in the model. Most

apparent, the parameters interpreted as outflow transition rates from the heavy user

category directly affect the estimates of cost-effectiveness of treatment programs and,

accordingly, the relative cost-effectiveness advantage cited by Controlling Cocaine for

treatment as compared to the supply control programs.

V-9

D. EXTRAPOLATION OF THE MARGINAL COST OF ADDITIONAL NEWTREATMENTS

Controlling Cocaine found that treatment programs are far more cost-effective

than are any of the supply control programs it considered, including source-zone

interdiction. Two assumptions appear to have played a major role in determining the cost-

effectiveness of treatment programs:

"* Assumption 1. The marginal cost of providing additional new treatmentswould not change from the cost data used in Controlling Cocaine.

"* Assumption 2. Increased funds directed at treatment programs would beexpended solely to increase the treatment of heavy users, that is, those who use

cocaine at least weekly, as opposed to light users who use cocaine at least once

a year but less than weekly.

Both of these key assumptions are prominently displayed in the Summary of

Controlling Cocaine, which states:

The average cocaine treatment ... costs $1,740 per person treated, so $34million pays for 19,500 treatments. These additional treatments areassumed to be given to heavy cocaine users ....

Since Controlling Cocaine was published in 1994, the amount of funding directed

toward treatment programs has increased substantially above inflation. Figure V-2 depicts

the funding (squares) for the majority of public treatments (approximately 75 percent) as

derived from the Veteran's Administration (VA) and DoHHS (SAMHSA) data, and the

number of public treatments (diamonds) actually delivered as extracted from The National

Drug Control Strategy, 1995 (ONDCP, 1995) and published VA reports (GAO, 1996).

Under Assumption 1, the number of treatments from 1989 to 1995 should have increased

significantly since an additional billion dollars was made available for treatment over that

period. Instead, the actual number of treatments declined by a few percent. A past

Director of ONDCP testified before Congress3 that "between 1988 and 1993, we roughly

tripled the treatment budget of the Federal Government," while the "number of people

treated per year declined." Since Controlling Cocaine assumed private treatments as 35

percent of the total treatments and public treatments as 65 percent of the total (page 88,

Table D-1, Note), this analysis has assumed the same ratio. Since the expenditures have

3 U.S. House of Representatives Report 104-486, "National Drug Policy: A Review of the Status of the Drug

War," March 1996.

V-10

Treatment Program History'i" 2.5C0

Q 1. . . .. .... .. ... . .. . . .. .. .. .. ..... . . . ... .. . ... . ... . . . .... .. . . . . .. ... . .. .. ... i . .. .. . .. .... . .. .. .. . . . . . . . ....

2-

01.5 - . . . .. . . . ... . . . .. . . . . ..C

.Q)

E

LL- 02

H--

-o

Xi~ 0 I I I I I

1989 1990 1991 1992 1993 1994 1995 1996Time (year)

Figure V-2. History of federally-funded drug treatments and funding expenditures. Opensquares are treatment expenditures in billions of dollars, open diamonds are number of

treatments delivered in millions.

approximately doubled in real terms and the number of treatments have slightly declined,

this suggests that the cost reported in Controlling Cocaine as $34 million to effect a onepercent reduction in consumption over 15 years is low.

Since this paper did not assess the effectiveness of counterdrug programs beyondinterdiction, no data relating to Assumption 2 are presented here. Still, it seems likely thatsome portion of the increased treatment funding has been expended on less-than-heavy

users. For example, in determining the proportion of heavy users who stop using cocainewhile undergoing treatment, Controlling Cocaine relies upon a 1989 study of a number of

users who underwent treatment the previous year. The footnote on page 88 of the report

states that of this study group, "12.8 percent of those treated were heavy users." In viewof the fact that the vast majority of those receiving treatment were not heavy users, it is

unlikely that all of the increases in treatment funding have been directed at heavy users.Since it is not apparent that all new expenditures could be targeted only to heavy users,

especially with declining user populations (according to the underlying actual data), thereported $34 million to effect a one percent reduction in consumption again appears to belower than the actual data indicates.

V-1I

E. INCLUSION OF INAPPROPRIATE COSTS IN DETERMINING THECOSTS OF SOURCE-COUNTRY CONTROL AND INTERDICTION

The determination of cost-effectiveness is significantly affected by the costs

determined as appropriate. Significant inconsistencies appear in the numbers adopted as

inputs to the cost-effectiveness calculations of Controlling Cocaine.

Controlling Cocaine bases its estimates of costs on the funds expended in 1992.

In general, Controlling Cocaine appears to overestimate costs for source-country control

and interdiction, based upon the actual funds expended in 1992 as reported by the

ONDCP (1996) report. For example, under the costs for interdiction, Controlling

Cocaine included state and local assistance provided by DoD ($189 million) and the DoD

research and development budget ($91 million). The largest portion of the funds in the

DoD R&D budget is for the development of advanced drug detection technologies to

support domestic law enforcement and, therefore, these costs should not have been

allocated to interdiction. Making these adjustments to the 1992 estimate of interdiction

costs reduces the costs for interdiction from the $1.7 billion reported in Controlling

Cocaine to about $1.5 billion.

Similar problems exist when comparing Controlling Cocaine's costs for source-

country control. Of particular note:

N ONDCP (1996) reports $161 million for "DEA international" in 1992 whileTable B.7 of Controlling Cocaine reports $461 million.

0 Controlling Cocaine includes the costs for the US Agency for InternationalDevelopment (USAID) as part of the costs for source-country control, whileoffering no argument as to how such humanitarian aid contributes to seizingcoca products, trafficker equipment, or arresting drug traffickers, or otherwisedirectly contributes to the interdiction of the flow of coca products. Webelieve funds for USAID should not be included as part of the costs forsource-country control.

Controlling Cocaine included all costs of the DEA except "domesticenforcement and state and local task forces." This portrays the DEA assomething other than a domestic law enforcement agency. ControllingCocaine implicitly charges 90 percent of all data processing costs, all state andlocal training costs, all administrative costs, all diversion control costs, allresearch and development costs, etc., to source-country control costs. Takinginto account that the DEA operates international offices in 42 countries andonly Peru, Colombia, and Bolivia are cocaine source nations, it seemsunreasonable that 90 percent of the international DEA budget should beallocated to cocaine source-country control as is done in Controlling Cocaine.

V-12

Making these adjustments and using the data reported for 1992 in ONDCP (1996),

source-country control has an approximate cost in 1992 of about $300 million versus the

$871 million reported in Controlling Cocaine.

F. SUMMARY OF COST-EFFECTIVENESS ESTIMATES

Figure V-3 compares the estimates of cost-effectiveness for three of the four

cocaine-control programs presented in Controlling Cocaine and those reported in thiswork. These are source-country control (source-zone interdiction in the terminology of

this paper); interdiction (overall); and treatment. This paper attempted no estimate of the

cost-effectiveness of domestic enforcement. Figure V-3 indicates significant differences

between the estimates of cost-effectiveness presented in Controlling Cocaine and those

developed in this paper.

783 366100.

Z Controlling Cocaine, 1994

E 3i This work

so 3os

t; -30-60

.. i>34

34

- 13-25

0 - -

Source Country Control Interdiction Treatment

Figure V-3. Comparison of program cost-effectiveness estimates.

V-13

In considering the estimates of this paper, the reader is cautioned that:

"The estimate of $13 million to $25 million per year for source-zone interdictionwas derived from an estimated expenditure range of $200 million to $400million per year for the period from 1992 to 1995. It is unlikely that all sourcezone actions have been equally cost-effective. Nonetheless, this empiricalresult does suggest that a series of well-conceived source-zone actions canproduce supply disruptions resulting in significant increases in the domesticstreet price of cocaine at relatively modest costs.

" The estimate of $30 million to $60 million per year for the overall interdictioneffort is based on a hypothesis that, in the absence of the War on Drugs, the

street price index would have continued to decline along the trend line evidentprior to 1989. While no detailed economic theory that prescribes how theprice of a monopolistic or oligopolistic commodity, under the pressure ofencroaching competition, approaches equilibrium is presented, the hypothesisis broadly consistent with such theories. Even without this hypothesis, it is stillpossible to argue, albeit only qualitatively (e.g., from Figures IV-4 and IV-5),that interdiction has caused a significant change in the rate of cocaine pricedecrease.

The "IDA estimate" of the cost-effectiveness of treatment is merely a simplisticadjustment to accommodate actual experience regarding the number oftreatments provided for the treatment funds expended. The authors of thispaper have not studied treatment programs in sufficient detail to assess theaccuracy of each of the data presented in the previous study other than tocheck data sources for the actual experience in order to correct underlyingassumptions.

In conclusion, the estimates presented in this paper indicate that each control

program reported here has a cost-effectiveness of the same order of magnitude. If each

control program is implemented effectively, the two approaches, treatment and

interdiction, are complementary - for a given level of funding, interdiction can have a

significant immediate response that largely vanishes when funding is terminated, 4 while

treatment, with only a small initial response, has effects that persist even after funding is

terminated. Both approaches will fail to have a permanent effect when funding is

terminated if the incidence rate is not reduced through other processes (e.g., education).

Ultimately, this analysis argues for a balanced approach to drug control that includes

interdiction, treatment, and incidence reduction programs.

4 To the degree that interdiction increases the price of cocaine, it also causes some reduction in incidence. Theseincidence reductions have a persisting effect over time.

V-14

CHAPTER VI

FINDINGS

VI. FINDINGS

Based upon the data and analysis presented in this paper, the following are IDA's

three principal findings:

1. Well-conceived source-zone operations, in cooperation with host nation

forces, that significantly and unexpectedly disrupt the normal drug traffickerprocesses for producing and transporting coca products from the source zone,cause discernible increases in the street price of cocaine in the U.S., and,through normal market relationships between price and demand, therebyreduce cocaine consumption.

2. Significant source-zone interdiction activities can produce a lasting increase inthe street price of cocaine in the U.S. resulting from higher costs oftransshipment.

3. Supply disruptions have significant effects on the street price of cocaine.Most other policy studies to date have assumed an "additive" structure andweak price effects; these assumptions are questionable on both empirical andtheoretical grounds.

Additional relevant findings are:

4. Despite previously published reports based on models of demand that reflectcontinuing increases, improved modeling and additional data both indicatethat the total number of cocaine users has been declining since approximately1985.

5. When the supply of cocaine to the market is obstructed, traffickers respond byreducing the purity of the cocaine sold in the U.S. in an apparent attempt tosupply customers at prices close to the customary price per dose.

6. STRIDE data can be used in combination with robust analysis techniques toassess the effectiveness of supply control activities and to provide feedbackuseful to those executing such activities.

7. The low cost effectiveness of interdiction as contrasted to the costeffectiveness of demand control, as previously reported in ControllingCocaine, is inconsistent with the available data.

8. Several indirect, but nonetheless logical, indicators of cocaine usage show

clear inverse correlation with STRIDE price excursions.

VI-1

These findings in no way suggest that any particular element of supply or demand

control is not a valuable component of a comprehensive strategy to counter drug use.

Rather, they demonstrate that the vast differences of cost-effectiveness concluded by

earlier work are not supported by the currently available data. Ultimately, our analysis

argues for a balanced approach to drug control that would include interdiction, treatment,

and incidence reduction programs.

VI-2

REFERENCES

Atkinson, Anthony B. and Stiglitz, Joseph E., Lectures on Public Economics, New York:McGraw-Hill, 1980.

Boyum, D.A., Reflections on Economic Theory and Drug Enforcement, Ph.D.Dissertation, Harvard University, 1992.

Caulkins, J., Developing a Price Series for Cocaine, RAND Drug Policy ResearchCenter, 1993

Caulkins, J., The Distribution and Consumption of Illicit Drugs: Some MathematicalModels and Their Policy Implications, Ph.D. Dissertation, Massachusetts Instituteof Technology, 1990.

Caulkins, JP. and Padman, R., "Quantity Discounts and Quality Premia for IllicitDrugs," J. Am. Stat. Assoc., Vol. 88 (1993): 748-757.

Cave, J.K. and Reuter, P., The Interdictor's Lot: A Dynamic Model of the Market forDrug Smuggling Services, RAND Corp. (N-2632-USDP), 1988.

CORAH, Produccion de la Coca en el Alto Huallaga y su Impacto Socio Economico,Lima, Peru, October 1994.

Cootner, P.H., The Random Character of Stock Market Prices, Cambridge: M.I.T. Press,1964.

Craig, J., "Market Has Slipped, but Farmers Find It Hard to Stop Growing Coca," Reuter,Washington Times, 15 September 1995.

Crane, B., Rivolo, A.R., and Comfort, G., An Empirical Examination of CounterdrugInterdiction Program Effectiveness, IDA Document D-1837 (draft), 1996.

Cuanta, S.A., Coca's Impact on Peru's Economy, 1980-1992, Lima, Peru, September1993.

DAWN, Drug Abuse Warning Network, Substance Abuse and Mental Health ServicesAdministration, Department of Health and Human Services.

DEA, 1994-1996, Drug Enforcement Administration, System to Retrieve Informationfrom Drug Evidence, monthly data, Department of Justice, Deputy Administratorfor Intelligence, Statistics Division.

DEA, 1996, Drug Enforcement Administration, Monthly Price and Purity Reports,Department of Justice, and private communication with the Statistics Division.

DiNardo, J., "Law Environment, the Price of Cocaine and Cocaine Use," Meth. ComputerModeling, Vol. 17, No. 2 (1993).

DUF, Drug Use Forecast, National Institute of Justice, Department of Justice

DuMouchel, W.H., "Stable Distributions in Statistical Inference: 1. Symmetric StableDistributions Compared to Other Symmetric Long-Tailed Distributions," J. Am.Stat. Assoc., Vol. 68, No. 342, (1973): 469-477.

R-1

Efron, B., "More Efficient Bootstrap Computations," J. Am. Stat. Assoc., Vol. 85, No.409, (March 1990).

Efron, B. and Gong, G., "A Leisurely Look at the Bootstrap, the Jackknife, and Cross-Validation," The Am. Statistician, Vol 37, No. 1 (1983).

Efron, B. and Tibshirani, R., "Statistical Data Analysis in the Computer Age," Science,Vol. 253 (26 July 1991): 390-395.

Everingham, S.S., and Rydell, C.P., Modeling the Demand for Cocaine, MR-332-ONCDP/A/DPRC, RAND Corp, 1994.

Everingham, S.S., and Rydell, C.P., and Caulkins, J.P., "Cocaine Consumption in theUnited States: Estimating Past Trends and Future Scenarios," Socio-Econ Plann.Sci, Vol 29, No. 4, (1995): 305-314.

Grossman, M., Chaloupka, F. and Brown, C., The Demand for Cocaine by Young Adults:A Rational Addiction Approach, 1996 (unpublished).

Hyatt, R.R., and W. Rhodes, "The Price and Purity of Cocaine: The Relationship toEmergency Room Visits and Death, and to Drug Use Among Arrestees,"Statistics in Medicine, Vol 14 (1995): 655-668.

Krauss, C., "Colombia Arrests Raise Price of Cocaine in New York City," The New YorkTimes, 15 September 1995.

Laity, J., The Coca Economy in the Upper Huallaga, Lima, Peru, 1989.

Mandelbrot, B.B., "Price Change and Scaling in Economics," The Fractal Geometry ofNature, W. H. Freeman and Co., 1983: 334-340.

Mandelbrot, B.B., "Stable Paretian Random Functions and the Multiplicative Variation ofIncome," Econometrica, Vol. 29, No. 4 (Oct. 1961): 517-543.

Moore, M.H, "Supply Reduction and Drug Law Enforcement," Drugs and Crime,M.Tonry and J. Q. Wilson (eds.), Vol. 13 of Crime and Justice: A Review ofResearch, Chicago: U. of Chicago Press, 1990.

Moore, M.H, "Policy Concerning Drug Abuse in New York State," Volume III:Economics of Heroin Distribution, HI-1354-FR, The Hudson Institute, 20 July1970.

Mosteller, F. and Tukey, J.W., Data Analysis and Regression, Addison-Wesley Pub. Co.,1977

NHSDA, National Household Survey on Drug Abuse, Substance Abuse and MentalHealth Services Administration, Department of Health and Human Services,1996.

Nunez, J. and R. Reategui, La Economia Cocalera en el Alto Huallaga: ImpactoEconomico, Tesis de Bachiller, Universidad del Pacifico, Lima, Peru, 1990.

ONDCP, National Drug Control Strategy, The White House, Washington DC, 1996.

ONDCP, National Drug Control Strategy (Table B-9), The White House, WashingtonDC, 1995.

ONDCP, National Drug Control Strategy Budget Summary, The White House,Washington DC, 1995.

R-2

PEAH, 1996, Proyecto Especial Alto Huallaga (PEAH/OSE - Special Project on theUpper Huallaga/Office of Sectoral Statistics) monthly price reports from Lima,Peru (through September 1996).

Preble and Casey, "Taking Care of Business - The Life of the Heroin Addict on theStreet," Int. J. of Addictions, Vol. IV, No. 1 (1967): 12.

Reuter, P., Crawford, G., and Cave, J., Sealing the Borders: The Effects of IncreasedMilitary Participation in Drug Interdiction, R-3594-USDP, RAND Corp., 1988.

Reuter, P., and Kleiman, M.A.R., Risks and Prices: An Economic Analysis of DrugEnforcement, University of Chicago, 1986.

Rhodes, W., et. al., What America's Users Spend on Illegal Drugs, 1988-1993, AbtAssociates Inc., 1995.

Riley, K.J., The Efficacy of Source Country Cocaine Policies, Ph.D. Dissertation, RGSD-102, RAND Corp., 1993.

Rydell, C.P., and Everingham, S.S., Controlling Cocaine: Supply Versus DemandPrograms, MR-331-ONDCP/A/DPRC, RAND Corp., 1994.

Rydell, C.P., Caulkins, J.P. and Everingham, S.S., "Enforcement of Treatment?Modeling the Relative Efficacy of Alternatives for Controlling Cocaine,"Operations Research, Vol. 44, No. 5, September-October 1996.

Saffer, H. and Chaloupka, F., The Demand for Illicit Drugs, National Bureau ofEconomic Research, 1996 (in preparation).

SAMHSA, Preliminary Estimates from the 1994 National Household Survey on DrugAbuse, Advance Report No. 10, U.S. Department of Health and Human Services,September 1995.

SBCL, SmithKline Beecham Clinical Laboratory positive test internal monthly data forcocaine, opiates, and amphetamines, October 1996.

TEDS, Treatment Episode Data Set, Substance Abuse and Mental Health ServicesAdministration, U.S. Department of Health and Human Services.

UNDCP, United Nations Drug Control Program .(UNDCP), Situation Analysis on the Fallof Coca Prices in Peru, 1995.

USAID, Upper Huallaga Agricultural Development, Project Paper, August, 1981.

USAID, Andean Counter-Drug Initiative, Objective IV: Sustainable Development and theCounter-Narcotics Strategy, Semi-Annual Report, April-September 1993 -December 1993.

U.S. Embassy, Peru, Memorandum OASD/SOLIC/DEP&S, Subject: PoliticalSustainment of Interdiction Success, 5 January 1996.

Weisman, A., "Can Colombia Finally Win the War on Drugs? Better to Ask: Can ItSurvive If It Does Win?" The Los Angeles Times, 24 September 1995.

Zabludof, S., "Economics and Narcotics: A Perspective," Economics of the NarcoticsIndustry, report of conference of Bureau of Intelligence and Research, U.S.Department of State, and Central Intelligence Agency, 21-22 November 1994.

R-3

APPENDIX A

ACRONYMS

APPENDIX ALIST OF ACRONYMS

AF Air ForceARPA Advanced Research Project AgencyDAWN Drug Abuse Warning NetworkDEA Drug Enforcement AdministrationDoD Department of DefenseDoHHS Department of Health and Human ServicesDUF Drug Usage ForecastGAO General Accounting OfficeIDA Institute for Defense AnalysesNHSDA National Household Survey on Drug AbuseONDCP Office of National Drug Control PolicyPEAH Projecto Especial Alto HuallagaSAMHSA Substance Abuse and Mental Health Services AdmistrationSBCL SmithKline Beecham Clinical LaboratoriesSJ Support JusticeSTRIDE System to Retrieve Information from Drug EvidenceTAT Tactical Assessment TeamTEDS Treatment Episode Data SetTIC The Interdiction CommitteeU.S. United StatesUNDCP United Nations Drug Control ProgramUSAID US Agency for International DevelopmentUSIC U.S. Interdiction CoordinatorMT Metric Ton

A-1

REPORT DOCUMENTATION PAGE Form Approved

I OMB No. 0704-0188Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering andmaintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,including suggestions for reducing this burden, to Washington Headquarters Services. Directorate for information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington.VA 22202-4302. and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188), Washington. DC 20503

1. AGENCY USE ONLY (Leave blank) 2. REPORT DATE 3. REPORT TYPE AND DATES COVERED

January 1997 Final; 1985 - 1996

4. TITLE AND SUBTITLE 5. FUNDING NUMBERS

An Empirical Examination of Counterdrug Interdiction Program DASW01-94-C-0054, OUSD(A)Effectiveness Task no. A-173

6. AUTHOR(S)

Barry D. Crane, A. Rex Rivolo, Gary C. Comfort

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION

REPORT NUMBERInstitute for Defense Analyses1801 N. Beauregard St. IDA Paper P-3219Alexandria, VA 22311-1772

9. SPONSORINGJMONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSORING/MONITORINGMr. Brian Sheridan, ASD(SO/LIC) AGENCY REPORT NUMBER

Deputy Assistant Secretary of Defense,Drug Enforcement Policy and SupportThe Pentagon, Rm. 2E258, Washington, DC 20301

11. SUPPLEMENTARY NOTES

12a. DISTRIBUTIONIAVAILABILITY STATEMENT 12b. DISTRIBUTION CODE

Approved for public release; distribution unlimited. Office of the DODCoordinator for Drug Enforcement Policy and Support, 28 January 1997.

13. ABSTRACT (Maximum 200 words)

Analysis of cocaine street prices since 1985, intelligence data regarding cocaine base movement in the Andeannations of South America, and cocaine usage indicators are used to arrive at the conclusion that a source zoneinterdiction strategy to disrupt significantly the production/transportation of coca base is a cost-effective

operational strategy for increasing cocaine prices, and thereby for reducing cocaine use in the United States.Empirical evidence supporting the correlation of source zone interdiction effectiveness with the price of cocainein the United States is presented, arguing that specific source zone operations since 1989 have directly affected

cocaine markets, resulting in price increases as large as 100 percent.

14. SUBJECT TERMS 15. NUMBER OF PAGES

cocaine, counter drug enforcement and interdiction 10416. PRICE CODE

17. SECURITY CLASSIFICATION 18. SECURITY CLASSIFICATION 19. SECURITY CLASSIFICATION 20. LIMITATION OFOF REPORT OF THIS PAGE OF ABSTRACT ABSTRACT

Unclassified Unclassified UnclassifiedNSN 7540-01-280-5500 Standard Form 298 (Rev. 2-89)

Prescribed by ANSI Std. Z39-18298-102

institute f'or defense analysesnonparametric statistical techniques are used to circumvent...

Documents